Upload folder using huggingface_hub

.gitattributes

@@ -18,3 +18,4 @@ sam/src/groundingdino/.asset/hero_figure.png filter=lfs diff=lfs merge=lfs -text
 tutorial/img/click_segment.jpg filter=lfs diff=lfs merge=lfs -text
 tutorial/img/input_video.jpg filter=lfs diff=lfs merge=lfs -text
 tutorial/img/start_tracking.jpg filter=lfs diff=lfs merge=lfs -text
+tool/weights/groundingdino_swint_ogc.pth filter=lfs diff=lfs merge=lfs -text

tool/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2023 - present, IDEA Research.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

tool/README.md

@@ -0,0 +1,367 @@
+<div align="center">
+  <img src="./.asset/grounding_dino_logo.png" width="30%">
+</div>
+
+# :sauropod: Grounding DINO 
+
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/grounding-dino-marrying-dino-with-grounded/zero-shot-object-detection-on-mscoco)](https://paperswithcode.com/sota/zero-shot-object-detection-on-mscoco?p=grounding-dino-marrying-dino-with-grounded) [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/grounding-dino-marrying-dino-with-grounded/zero-shot-object-detection-on-odinw)](https://paperswithcode.com/sota/zero-shot-object-detection-on-odinw?p=grounding-dino-marrying-dino-with-grounded) \
+[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/grounding-dino-marrying-dino-with-grounded/object-detection-on-coco-minival)](https://paperswithcode.com/sota/object-detection-on-coco-minival?p=grounding-dino-marrying-dino-with-grounded) [![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/grounding-dino-marrying-dino-with-grounded/object-detection-on-coco)](https://paperswithcode.com/sota/object-detection-on-coco?p=grounding-dino-marrying-dino-with-grounded)
+
+
+**[IDEA-CVR, IDEA-Research](https://github.com/IDEA-Research)** 
+
+[Shilong Liu](http://www.lsl.zone/), [Zhaoyang Zeng](https://scholar.google.com/citations?user=U_cvvUwAAAAJ&hl=zh-CN&oi=ao), [Tianhe Ren](https://rentainhe.github.io/), [Feng Li](https://scholar.google.com/citations?user=ybRe9GcAAAAJ&hl=zh-CN), [Hao Zhang](https://scholar.google.com/citations?user=B8hPxMQAAAAJ&hl=zh-CN), [Jie Yang](https://github.com/yangjie-cv), [Chunyuan Li](https://scholar.google.com/citations?user=Zd7WmXUAAAAJ&hl=zh-CN&oi=ao), [Jianwei Yang](https://jwyang.github.io/), [Hang Su](https://scholar.google.com/citations?hl=en&user=dxN1_X0AAAAJ&view_op=list_works&sortby=pubdate), [Jun Zhu](https://scholar.google.com/citations?hl=en&user=axsP38wAAAAJ), [Lei Zhang](https://www.leizhang.org/)<sup>:email:</sup>.
+
+
+[[`Paper`](https://arxiv.org/abs/2303.05499)] [[`Demo`](https://huggingface.co/spaces/ShilongLiu/Grounding_DINO_demo)] [[`BibTex`](#black_nib-citation)]
+
+
+PyTorch implementation and pretrained models for Grounding DINO. For details, see the paper **[Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection](https://arxiv.org/abs/2303.05499)**.
+
+## :sun_with_face: Helpful Tutorial
+
+- :grapes: [[Read our arXiv Paper](https://arxiv.org/abs/2303.05499)]
+- :apple:  [[Watch our simple introduction video on YouTube](https://youtu.be/wxWDt5UiwY8)]
+- :blossom:   &nbsp;[[Try the Colab Demo](https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/zero-shot-object-detection-with-grounding-dino.ipynb)]
+- :sunflower: [[Try our Official Huggingface Demo](https://huggingface.co/spaces/ShilongLiu/Grounding_DINO_demo)]
+- :maple_leaf: [[Watch the Step by Step Tutorial about GroundingDINO by Roboflow AI](https://youtu.be/cMa77r3YrDk)]
+- :mushroom: [[GroundingDINO: Automated Dataset Annotation and Evaluation by Roboflow AI](https://youtu.be/C4NqaRBz_Kw)]
+- :hibiscus: [[Accelerate Image Annotation with SAM and GroundingDINO by Roboflow AI](https://youtu.be/oEQYStnF2l8)]
+- :white_flower: [[Autodistill: Train YOLOv8 with ZERO Annotations based on Grounding-DINO and Grounded-SAM by Roboflow AI](https://github.com/autodistill/autodistill)]
+
+<!-- Grounding DINO Methods | 
+[![arXiv](https://img.shields.io/badge/arXiv-2303.05499-b31b1b.svg)](https://arxiv.org/abs/2303.05499) 
+[![YouTube](https://badges.aleen42.com/src/youtube.svg)](https://youtu.be/wxWDt5UiwY8) -->
+
+<!-- Grounding DINO Demos |
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/zero-shot-object-detection-with-grounding-dino.ipynb) -->
+<!-- [![YouTube](https://badges.aleen42.com/src/youtube.svg)](https://youtu.be/cMa77r3YrDk)
+[![HuggingFace space](https://img.shields.io/badge/🤗-HuggingFace%20Space-cyan.svg)](https://huggingface.co/spaces/ShilongLiu/Grounding_DINO_demo)
+[![YouTube](https://badges.aleen42.com/src/youtube.svg)](https://youtu.be/oEQYStnF2l8)
+[![YouTube](https://badges.aleen42.com/src/youtube.svg)](https://youtu.be/C4NqaRBz_Kw) -->
+
+## :sparkles: Highlight Projects
+
+- [Semantic-SAM: a universal image segmentation model to enable segment and recognize anything at any desired granularity.](https://github.com/UX-Decoder/Semantic-SAM), 
+- [DetGPT: Detect What You Need via Reasoning](https://github.com/OptimalScale/DetGPT)
+- [Grounded-SAM: Marrying Grounding DINO with Segment Anything](https://github.com/IDEA-Research/Grounded-Segment-Anything)
+- [Grounding DINO with Stable Diffusion](demo/image_editing_with_groundingdino_stablediffusion.ipynb)
+- [Grounding DINO with GLIGEN for Controllable Image Editing](demo/image_editing_with_groundingdino_gligen.ipynb)
+- [OpenSeeD: A Simple and Strong Openset Segmentation Model](https://github.com/IDEA-Research/OpenSeeD)
+- [SEEM: Segment Everything Everywhere All at Once](https://github.com/UX-Decoder/Segment-Everything-Everywhere-All-At-Once)
+- [X-GPT: Conversational Visual Agent supported by X-Decoder](https://github.com/microsoft/X-Decoder/tree/xgpt)
+- [GLIGEN: Open-Set Grounded Text-to-Image Generation](https://github.com/gligen/GLIGEN)
+- [LLaVA: Large Language and Vision Assistant](https://github.com/haotian-liu/LLaVA)
+
+<!-- Extensions | [Grounding DINO with Segment Anything](https://github.com/IDEA-Research/Grounded-Segment-Anything); [Grounding DINO with Stable Diffusion](demo/image_editing_with_groundingdino_stablediffusion.ipynb); [Grounding DINO with GLIGEN](demo/image_editing_with_groundingdino_gligen.ipynb)  -->
+
+
+
+<!-- Official PyTorch implementation of [Grounding DINO](https://arxiv.org/abs/2303.05499), a stronger open-set object detector. Code is available now! -->
+
+
+## :bulb: Highlight
+
+- **Open-Set Detection.** Detect **everything** with language!
+- **High Performancce.** COCO zero-shot **52.5 AP** (training without COCO data!). COCO fine-tune **63.0 AP**.
+- **Flexible.** Collaboration with Stable Diffusion for Image Editting.
+
+
+
+
+## :fire: News
+- **`2023/07/18`**: We release [Semantic-SAM](https://github.com/UX-Decoder/Semantic-SAM), a universal image segmentation model to enable segment and recognize anything at any desired granularity. **Code** and **checkpoint** are available!
+- **`2023/06/17`**: We provide an example to evaluate Grounding DINO on COCO zero-shot performance.
+- **`2023/04/15`**: Refer to [CV in the Wild Readings](https://github.com/Computer-Vision-in-the-Wild/CVinW_Readings) for those who are interested in open-set recognition!
+- **`2023/04/08`**: We release [demos](demo/image_editing_with_groundingdino_gligen.ipynb) to combine [Grounding DINO](https://arxiv.org/abs/2303.05499) with [GLIGEN](https://github.com/gligen/GLIGEN)  for more controllable image editings.
+- **`2023/04/08`**: We release [demos](demo/image_editing_with_groundingdino_stablediffusion.ipynb) to combine [Grounding DINO](https://arxiv.org/abs/2303.05499) with [Stable Diffusion](https://github.com/Stability-AI/StableDiffusion) for image editings.
+- **`2023/04/06`**: We build a new demo by marrying GroundingDINO with [Segment-Anything](https://github.com/facebookresearch/segment-anything) named **[Grounded-Segment-Anything](https://github.com/IDEA-Research/Grounded-Segment-Anything)** aims to support segmentation in GroundingDINO.
+- **`2023/03/28`**: A YouTube [video](https://youtu.be/cMa77r3YrDk) about Grounding DINO and basic object detection prompt engineering. [[SkalskiP](https://github.com/SkalskiP)]
+- **`2023/03/28`**: Add a [demo](https://huggingface.co/spaces/ShilongLiu/Grounding_DINO_demo) on Hugging Face Space!
+- **`2023/03/27`**: Support CPU-only mode. Now the model can run on machines without GPUs.
+- **`2023/03/25`**: A [demo](https://colab.research.google.com/github/roboflow-ai/notebooks/blob/main/notebooks/zero-shot-object-detection-with-grounding-dino.ipynb) for Grounding DINO is available at Colab. [[SkalskiP](https://github.com/SkalskiP)]
+- **`2023/03/22`**: Code is available Now!
+
+<details open>
+<summary><font size="4">
+Description
+</font></summary>
+ <a href="https://arxiv.org/abs/2303.05499">Paper</a> introduction.
+<img src=".asset/hero_figure.png" alt="ODinW" width="100%">
+Marrying <a href="https://github.com/IDEA-Research/GroundingDINO">Grounding DINO</a> and <a href="https://github.com/gligen/GLIGEN">GLIGEN</a>
+<img src="https://huggingface.co/ShilongLiu/GroundingDINO/resolve/main/GD_GLIGEN.png" alt="gd_gligen" width="100%">
+</details>
+
+## :star: Explanations/Tips for Grounding DINO Inputs and Outputs
+- Grounding DINO accepts an `(image, text)` pair as inputs.
+- It outputs `900` (by default) object boxes. Each box has similarity scores across all input words. (as shown in Figures below.)
+- We defaultly choose the boxes whose highest similarities are higher than a `box_threshold`.
+- We extract the words whose similarities are higher than the `text_threshold` as predicted labels.
+- If you want to obtain objects of specific phrases, like the `dogs` in the sentence `two dogs with a stick.`, you can select the boxes with highest text similarities with `dogs` as final outputs. 
+- Note that each word can be split to **more than one** tokens with different tokenlizers. The number of words in a sentence may not equal to the number of text tokens.
+- We suggest separating different category names with `.` for Grounding DINO.
+![model_explain1](.asset/model_explan1.PNG)
+![model_explain2](.asset/model_explan2.PNG)
+
+## :label: TODO 
+
+- [x] Release inference code and demo.
+- [x] Release checkpoints.
+- [x] Grounding DINO with Stable Diffusion and GLIGEN demos.
+- [ ] Release training codes.
+
+## :hammer_and_wrench: Install 
+
+**Note:**
+
+0. If you have a CUDA environment, please make sure the environment variable `CUDA_HOME` is set. It will be compiled under CPU-only mode if no CUDA available.
+
+Please make sure following the installation steps strictly, otherwise the program may produce: 
+```bash
+NameError: name '_C' is not defined
+```
+
+If this happened, please reinstalled the groundingDINO by reclone the git and do all the installation steps again.
+ 
+#### how to check cuda:
+```bash
+echo $CUDA_HOME
+```
+If it print nothing, then it means you haven't set up the path/
+
+Run this so the environment variable will be set under current shell. 
+```bash
+export CUDA_HOME=/path/to/cuda-11.3
+```
+
+Notice the version of cuda should be aligned with your CUDA runtime, for there might exists multiple cuda at the same time. 
+
+If you want to set the CUDA_HOME permanently, store it using:
+
+```bash
+echo 'export CUDA_HOME=/path/to/cuda' >> ~/.bashrc
+```
+after that, source the bashrc file and check CUDA_HOME:
+```bash
+source ~/.bashrc
+echo $CUDA_HOME
+```
+
+In this example, /path/to/cuda-11.3 should be replaced with the path where your CUDA toolkit is installed. You can find this by typing **which nvcc** in your terminal:
+
+For instance, 
+if the output is /usr/local/cuda/bin/nvcc, then:
+```bash
+export CUDA_HOME=/usr/local/cuda
+```
+**Installation:**
+
+1.Clone the GroundingDINO repository from GitHub.
+
+```bash
+git clone https://github.com/IDEA-Research/GroundingDINO.git
+```
+
+2. Change the current directory to the GroundingDINO folder.
+
+```bash
+cd GroundingDINO/
+```
+
+3. Install the required dependencies in the current directory.
+
+```bash
+pip install -e .
+```
+
+4. Download pre-trained model weights.
+
+```bash
+mkdir weights
+cd weights
+wget -q https://github.com/IDEA-Research/GroundingDINO/releases/download/v0.1.0-alpha/groundingdino_swint_ogc.pth
+cd ..
+```
+
+## :arrow_forward: Demo
+Check your GPU ID (only if you're using a GPU)
+
+```bash
+nvidia-smi
+```
+Replace `{GPU ID}`, `image_you_want_to_detect.jpg`, and `"dir you want to save the output"` with appropriate values in the following command
+```bash
+CUDA_VISIBLE_DEVICES={GPU ID} python demo/inference_on_a_image.py \
+-c groundingdino/config/GroundingDINO_SwinT_OGC.py \
+-p weights/groundingdino_swint_ogc.pth \
+-i image_you_want_to_detect.jpg \
+-o "dir you want to save the output" \
+-t "chair"
+ [--cpu-only] # open it for cpu mode
+```
+
+If you would like to specify the phrases to detect, here is a demo:
+```bash
+CUDA_VISIBLE_DEVICES={GPU ID} python demo/inference_on_a_image.py \
+-c groundingdino/config/GroundingDINO_SwinT_OGC.py \
+-p ./groundingdino_swint_ogc.pth \
+-i .asset/cat_dog.jpeg \
+-o logs/1111 \
+-t "There is a cat and a dog in the image ." \
+--token_spans "[[[9, 10], [11, 14]], [[19, 20], [21, 24]]]"
+ [--cpu-only] # open it for cpu mode
+```
+The token_spans specify the start and end positions of a phrases. For example, the first phrase is `[[9, 10], [11, 14]]`. `"There is a cat and a dog in the image ."[9:10] = 'a'`, `"There is a cat and a dog in the image ."[11:14] = 'cat'`. Hence it refers to the phrase `a cat` . Similarly, the `[[19, 20], [21, 24]]` refers to the phrase `a dog`.
+
+See the `demo/inference_on_a_image.py` for more details.
+
+**Running with Python:**
+
+```python
+from groundingdino.util.inference import load_model, load_image, predict, annotate
+import cv2
+
+model = load_model("groundingdino/config/GroundingDINO_SwinT_OGC.py", "weights/groundingdino_swint_ogc.pth")
+IMAGE_PATH = "weights/dog-3.jpeg"
+TEXT_PROMPT = "chair . person . dog ."
+BOX_TRESHOLD = 0.35
+TEXT_TRESHOLD = 0.25
+
+image_source, image = load_image(IMAGE_PATH)
+
+boxes, logits, phrases = predict(
+    model=model,
+    image=image,
+    caption=TEXT_PROMPT,
+    box_threshold=BOX_TRESHOLD,
+    text_threshold=TEXT_TRESHOLD
+)
+
+annotated_frame = annotate(image_source=image_source, boxes=boxes, logits=logits, phrases=phrases)
+cv2.imwrite("annotated_image.jpg", annotated_frame)
+```
+**Web UI**
+
+We also provide a demo code to integrate Grounding DINO with Gradio Web UI. See the file `demo/gradio_app.py` for more details.
+
+**Notebooks**
+
+- We release [demos](demo/image_editing_with_groundingdino_gligen.ipynb) to combine [Grounding DINO](https://arxiv.org/abs/2303.05499) with [GLIGEN](https://github.com/gligen/GLIGEN)  for more controllable image editings.
+- We release [demos](demo/image_editing_with_groundingdino_stablediffusion.ipynb) to combine [Grounding DINO](https://arxiv.org/abs/2303.05499) with [Stable Diffusion](https://github.com/Stability-AI/StableDiffusion) for image editings.
+
+## COCO Zero-shot Evaluations
+
+We provide an example to evaluate Grounding DINO zero-shot performance on COCO. The results should be **48.5**.
+
+```bash
+CUDA_VISIBLE_DEVICES=0 \
+python demo/test_ap_on_coco.py \
+ -c groundingdino/config/GroundingDINO_SwinT_OGC.py \
+ -p weights/groundingdino_swint_ogc.pth \
+ --anno_path /path/to/annoataions/ie/instances_val2017.json \
+ --image_dir /path/to/imagedir/ie/val2017
+```
+
+
+## :luggage: Checkpoints
+
+<!-- insert a table -->
+<table>
+  <thead>
+    <tr style="text-align: right;">
+      <th></th>
+      <th>name</th>
+      <th>backbone</th>
+      <th>Data</th>
+      <th>box AP on COCO</th>
+      <th>Checkpoint</th>
+      <th>Config</th>
+    </tr>
+  </thead>
+  <tbody>
+    <tr>
+      <th>1</th>
+      <td>GroundingDINO-T</td>
+      <td>Swin-T</td>
+      <td>O365,GoldG,Cap4M</td>
+      <td>48.4 (zero-shot) / 57.2 (fine-tune)</td>
+      <td><a href="https://github.com/IDEA-Research/GroundingDINO/releases/download/v0.1.0-alpha/groundingdino_swint_ogc.pth">GitHub link</a> | <a href="https://huggingface.co/ShilongLiu/GroundingDINO/resolve/main/groundingdino_swint_ogc.pth">HF link</a></td>
+      <td><a href="https://github.com/IDEA-Research/GroundingDINO/blob/main/groundingdino/config/GroundingDINO_SwinT_OGC.py">link</a></td>
+    </tr>
+    <tr>
+      <th>2</th>
+      <td>GroundingDINO-B</td>
+      <td>Swin-B</td>
+      <td>COCO,O365,GoldG,Cap4M,OpenImage,ODinW-35,RefCOCO</td>
+      <td>56.7 </td>
+      <td><a href="https://github.com/IDEA-Research/GroundingDINO/releases/download/v0.1.0-alpha2/groundingdino_swinb_cogcoor.pth">GitHub link</a>  | <a href="https://huggingface.co/ShilongLiu/GroundingDINO/resolve/main/groundingdino_swinb_cogcoor.pth">HF link</a> 
+      <td><a href="https://github.com/IDEA-Research/GroundingDINO/blob/main/groundingdino/config/GroundingDINO_SwinB.cfg.py">link</a></td>
+    </tr>
+  </tbody>
+</table>
+
+## :medal_military: Results
+
+<details open>
+<summary><font size="4">
+COCO Object Detection Results
+</font></summary>
+<img src=".asset/COCO.png" alt="COCO" width="100%">
+</details>
+
+<details open>
+<summary><font size="4">
+ODinW Object Detection Results
+</font></summary>
+<img src=".asset/ODinW.png" alt="ODinW" width="100%">
+</details>
+
+<details open>
+<summary><font size="4">
+Marrying Grounding DINO with <a href="https://github.com/Stability-AI/StableDiffusion">Stable Diffusion</a> for Image Editing
+</font></summary>
+See our example <a href="https://github.com/IDEA-Research/GroundingDINO/blob/main/demo/image_editing_with_groundingdino_stablediffusion.ipynb">notebook</a> for more details.
+<img src=".asset/GD_SD.png" alt="GD_SD" width="100%">
+</details>
+
+
+<details open>
+<summary><font size="4">
+Marrying Grounding DINO with <a href="https://github.com/gligen/GLIGEN">GLIGEN</a> for more Detailed Image Editing.
+</font></summary>
+See our example <a href="https://github.com/IDEA-Research/GroundingDINO/blob/main/demo/image_editing_with_groundingdino_gligen.ipynb">notebook</a> for more details.
+<img src=".asset/GD_GLIGEN.png" alt="GD_GLIGEN" width="100%">
+</details>
+
+## :sauropod: Model: Grounding DINO
+
+Includes: a text backbone, an image backbone, a feature enhancer, a language-guided query selection, and a cross-modality decoder.
+
+![arch](.asset/arch.png)
+
+
+## :hearts: Acknowledgement
+
+Our model is related to [DINO](https://github.com/IDEA-Research/DINO) and [GLIP](https://github.com/microsoft/GLIP). Thanks for their great work!
+
+We also thank great previous work including DETR, Deformable DETR, SMCA, Conditional DETR, Anchor DETR, Dynamic DETR, DAB-DETR, DN-DETR, etc. More related work are available at [Awesome Detection Transformer](https://github.com/IDEACVR/awesome-detection-transformer). A new toolbox [detrex](https://github.com/IDEA-Research/detrex) is available as well.
+
+Thanks [Stable Diffusion](https://github.com/Stability-AI/StableDiffusion) and [GLIGEN](https://github.com/gligen/GLIGEN) for their awesome models.
+
+
+## :black_nib: Citation
+
+If you find our work helpful for your research, please consider citing the following BibTeX entry.   
+
+```bibtex
+@article{liu2023grounding,
+  title={Grounding dino: Marrying dino with grounded pre-training for open-set object detection},
+  author={Liu, Shilong and Zeng, Zhaoyang and Ren, Tianhe and Li, Feng and Zhang, Hao and Yang, Jie and Li, Chunyuan and Yang, Jianwei and Su, Hang and Zhu, Jun and others},
+  journal={arXiv preprint arXiv:2303.05499},
+  year={2023}
+}
+```
+
+
+
+

tool/demo/create_coco_dataset.py

@@ -0,0 +1,83 @@
+import typer
+from groundingdino.util.inference import load_model, load_image, predict
+from tqdm import tqdm
+import torchvision
+import torch
+import fiftyone as fo
+
+
+def main(
+        image_directory: str = 'test_grounding_dino',
+        text_prompt: str = 'bus, car',
+        box_threshold: float = 0.15, 
+        text_threshold: float = 0.10,
+        export_dataset: bool = False,
+        view_dataset: bool = False,
+        export_annotated_images: bool = True,
+        weights_path : str = "groundingdino_swint_ogc.pth",
+        config_path: str = "../../GroundingDINO/groundingdino/config/GroundingDINO_SwinT_OGC.py",
+        subsample: int = None,
+    ):
+
+    model = load_model(config_path, weights_path)
+    
+    dataset = fo.Dataset.from_images_dir(image_directory)
+
+    samples = []
+
+    if subsample is not None: 
+        
+        if subsample < len(dataset):
+            dataset = dataset.take(subsample).clone()
+    
+    for sample in tqdm(dataset):
+
+        image_source, image = load_image(sample.filepath)
+
+        boxes, logits, phrases = predict(
+            model=model, 
+            image=image, 
+            caption=text_prompt, 
+            box_threshold=box_threshold, 
+            text_threshold=text_threshold,
+        )
+
+        detections = [] 
+
+        for box, logit, phrase in zip(boxes, logits, phrases):
+
+            rel_box = torchvision.ops.box_convert(box, 'cxcywh', 'xywh')
+
+            detections.append(
+                fo.Detection(
+                    label=phrase, 
+                    bounding_box=rel_box,
+                    confidence=logit,
+            ))
+
+        # Store detections in a field name of your choice
+        sample["detections"] = fo.Detections(detections=detections)
+        sample.save()
+
+    # loads the voxel fiftyone UI ready for viewing the dataset.
+    if view_dataset:
+        session = fo.launch_app(dataset)
+        session.wait()
+        
+    # exports COCO dataset ready for training
+    if export_dataset:
+        dataset.export(
+            'coco_dataset',
+            dataset_type=fo.types.COCODetectionDataset,
+        )
+        
+    # saves bounding boxes plotted on the input images to disk
+    if export_annotated_images:
+        dataset.draw_labels(
+            'images_with_bounding_boxes',
+            label_fields=['detections']
+        )
+
+
+if __name__ == '__main__':
+    typer.run(main)

tool/demo/gradio_app.py

@@ -0,0 +1,125 @@
+import argparse
+from functools import partial
+import cv2
+import requests
+import os
+from io import BytesIO
+from PIL import Image
+import numpy as np
+from pathlib import Path
+
+
+import warnings
+
+import torch
+
+# prepare the environment
+os.system("python setup.py build develop --user")
+os.system("pip install packaging==21.3")
+os.system("pip install gradio")
+
+
+warnings.filterwarnings("ignore")
+
+import gradio as gr
+
+from groundingdino.models import build_model
+from groundingdino.util.slconfig import SLConfig
+from groundingdino.util.utils import clean_state_dict
+from groundingdino.util.inference import annotate, load_image, predict
+import groundingdino.datasets.transforms as T
+
+from huggingface_hub import hf_hub_download
+
+
+
+# Use this command for evaluate the Grounding DINO model
+config_file = "groundingdino/config/GroundingDINO_SwinT_OGC.py"
+ckpt_repo_id = "ShilongLiu/GroundingDINO"
+ckpt_filenmae = "groundingdino_swint_ogc.pth"
+
+
+def load_model_hf(model_config_path, repo_id, filename, device='cpu'):
+    args = SLConfig.fromfile(model_config_path) 
+    model = build_model(args)
+    args.device = device
+
+    cache_file = hf_hub_download(repo_id=repo_id, filename=filename)
+    checkpoint = torch.load(cache_file, map_location='cpu')
+    log = model.load_state_dict(clean_state_dict(checkpoint['model']), strict=False)
+    print("Model loaded from {} \n => {}".format(cache_file, log))
+    _ = model.eval()
+    return model    
+
+def image_transform_grounding(init_image):
+    transform = T.Compose([
+        T.RandomResize([800], max_size=1333),
+        T.ToTensor(),
+        T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
+    ])
+    image, _ = transform(init_image, None) # 3, h, w
+    return init_image, image
+
+def image_transform_grounding_for_vis(init_image):
+    transform = T.Compose([
+        T.RandomResize([800], max_size=1333),
+    ])
+    image, _ = transform(init_image, None) # 3, h, w
+    return image
+
+model = load_model_hf(config_file, ckpt_repo_id, ckpt_filenmae)
+
+def run_grounding(input_image, grounding_caption, box_threshold, text_threshold):
+    init_image = input_image.convert("RGB")
+    original_size = init_image.size
+
+    _, image_tensor = image_transform_grounding(init_image)
+    image_pil: Image = image_transform_grounding_for_vis(init_image)
+
+    # run grounidng
+    boxes, logits, phrases = predict(model, image_tensor, grounding_caption, box_threshold, text_threshold, device='cpu')
+    annotated_frame = annotate(image_source=np.asarray(image_pil), boxes=boxes, logits=logits, phrases=phrases)
+    image_with_box = Image.fromarray(cv2.cvtColor(annotated_frame, cv2.COLOR_BGR2RGB))
+
+
+    return image_with_box
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser("Grounding DINO demo", add_help=True)
+    parser.add_argument("--debug", action="store_true", help="using debug mode")
+    parser.add_argument("--share", action="store_true", help="share the app")
+    args = parser.parse_args()
+
+    block = gr.Blocks().queue()
+    with block:
+        gr.Markdown("# [Grounding DINO](https://github.com/IDEA-Research/GroundingDINO)")
+        gr.Markdown("### Open-World Detection with Grounding DINO")
+
+        with gr.Row():
+            with gr.Column():
+                input_image = gr.Image(source='upload', type="pil")
+                grounding_caption = gr.Textbox(label="Detection Prompt")
+                run_button = gr.Button(label="Run")
+                with gr.Accordion("Advanced options", open=False):
+                    box_threshold = gr.Slider(
+                        label="Box Threshold", minimum=0.0, maximum=1.0, value=0.25, step=0.001
+                    )
+                    text_threshold = gr.Slider(
+                        label="Text Threshold", minimum=0.0, maximum=1.0, value=0.25, step=0.001
+                    )
+
+            with gr.Column():
+                gallery = gr.outputs.Image(
+                    type="pil",
+                    # label="grounding results"
+                ).style(full_width=True, full_height=True)
+                # gallery = gr.Gallery(label="Generated images", show_label=False).style(
+                #         grid=[1], height="auto", container=True, full_width=True, full_height=True)
+
+        run_button.click(fn=run_grounding, inputs=[
+                        input_image, grounding_caption, box_threshold, text_threshold], outputs=[gallery])
+
+
+    block.launch(server_name='0.0.0.0', server_port=7579, debug=args.debug, share=args.share)
+

tool/demo/inference_on_a_image.py

@@ -0,0 +1,214 @@
+import argparse
+import os
+import sys
+
+import numpy as np
+import torch
+from PIL import Image, ImageDraw, ImageFont
+
+import groundingdino.datasets.transforms as T
+from groundingdino.models import build_model
+from groundingdino.util import box_ops
+from groundingdino.util.slconfig import SLConfig
+from groundingdino.util.utils import clean_state_dict, get_phrases_from_posmap
+from groundingdino.util.vl_utils import create_positive_map_from_span
+
+
+def plot_boxes_to_image(image_pil, tgt):
+    H, W = tgt["size"]
+    boxes = tgt["boxes"]
+    labels = tgt["labels"]
+    assert len(boxes) == len(labels), "boxes and labels must have same length"
+
+    draw = ImageDraw.Draw(image_pil)
+    mask = Image.new("L", image_pil.size, 0)
+    mask_draw = ImageDraw.Draw(mask)
+
+    # draw boxes and masks
+    for box, label in zip(boxes, labels):
+        # from 0..1 to 0..W, 0..H
+        box = box * torch.Tensor([W, H, W, H])
+        # from xywh to xyxy
+        box[:2] -= box[2:] / 2
+        box[2:] += box[:2]
+        # random color
+        color = tuple(np.random.randint(0, 255, size=3).tolist())
+        # draw
+        x0, y0, x1, y1 = box
+        x0, y0, x1, y1 = int(x0), int(y0), int(x1), int(y1)
+
+        draw.rectangle([x0, y0, x1, y1], outline=color, width=6)
+        # draw.text((x0, y0), str(label), fill=color)
+
+        font = ImageFont.load_default()
+        if hasattr(font, "getbbox"):
+            bbox = draw.textbbox((x0, y0), str(label), font)
+        else:
+            w, h = draw.textsize(str(label), font)
+            bbox = (x0, y0, w + x0, y0 + h)
+        # bbox = draw.textbbox((x0, y0), str(label))
+        draw.rectangle(bbox, fill=color)
+        draw.text((x0, y0), str(label), fill="white")
+
+        mask_draw.rectangle([x0, y0, x1, y1], fill=255, width=6)
+
+    return image_pil, mask
+
+
+def load_image(image_path):
+    # load image
+    image_pil = Image.open(image_path).convert("RGB")  # load image
+
+    transform = T.Compose(
+        [
+            T.RandomResize([800], max_size=1333),
+            T.ToTensor(),
+            T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
+        ]
+    )
+    image, _ = transform(image_pil, None)  # 3, h, w
+    return image_pil, image
+
+
+def load_model(model_config_path, model_checkpoint_path, cpu_only=False):
+    args = SLConfig.fromfile(model_config_path)
+    args.device = "cuda" if not cpu_only else "cpu"
+    model = build_model(args)
+    checkpoint = torch.load(model_checkpoint_path, map_location="cpu")
+    load_res = model.load_state_dict(clean_state_dict(checkpoint["model"]), strict=False)
+    print(load_res)
+    _ = model.eval()
+    return model
+
+
+def get_grounding_output(model, image, caption, box_threshold, text_threshold=None, with_logits=True, cpu_only=False, token_spans=None):
+    assert text_threshold is not None or token_spans is not None, "text_threshould and token_spans should not be None at the same time!"
+    caption = caption.lower()
+    caption = caption.strip()
+    if not caption.endswith("."):
+        caption = caption + "."
+    device = "cuda" if not cpu_only else "cpu"
+    model = model.to(device)
+    image = image.to(device)
+    with torch.no_grad():
+        outputs = model(image[None], captions=[caption])
+    logits = outputs["pred_logits"].sigmoid()[0]  # (nq, 256)
+    boxes = outputs["pred_boxes"][0]  # (nq, 4)
+
+    # filter output
+    if token_spans is None:
+        logits_filt = logits.cpu().clone()
+        boxes_filt = boxes.cpu().clone()
+        filt_mask = logits_filt.max(dim=1)[0] > box_threshold
+        logits_filt = logits_filt[filt_mask]  # num_filt, 256
+        boxes_filt = boxes_filt[filt_mask]  # num_filt, 4
+
+        # get phrase
+        tokenlizer = model.tokenizer
+        tokenized = tokenlizer(caption)
+        # build pred
+        pred_phrases = []
+        for logit, box in zip(logits_filt, boxes_filt):
+            pred_phrase = get_phrases_from_posmap(logit > text_threshold, tokenized, tokenlizer)
+            if with_logits:
+                pred_phrases.append(pred_phrase + f"({str(logit.max().item())[:4]})")
+            else:
+                pred_phrases.append(pred_phrase)
+    else:
+        # given-phrase mode
+        positive_maps = create_positive_map_from_span(
+            model.tokenizer(text_prompt),
+            token_span=token_spans
+        ).to(image.device) # n_phrase, 256
+
+        logits_for_phrases = positive_maps @ logits.T # n_phrase, nq
+        all_logits = []
+        all_phrases = []
+        all_boxes = []
+        for (token_span, logit_phr) in zip(token_spans, logits_for_phrases):
+            # get phrase
+            phrase = ' '.join([caption[_s:_e] for (_s, _e) in token_span])
+            # get mask
+            filt_mask = logit_phr > box_threshold
+            # filt box
+            all_boxes.append(boxes[filt_mask])
+            # filt logits
+            all_logits.append(logit_phr[filt_mask])
+            if with_logits:
+                logit_phr_num = logit_phr[filt_mask]
+                all_phrases.extend([phrase + f"({str(logit.item())[:4]})" for logit in logit_phr_num])
+            else:
+                all_phrases.extend([phrase for _ in range(len(filt_mask))])
+        boxes_filt = torch.cat(all_boxes, dim=0).cpu()
+        pred_phrases = all_phrases
+
+
+    return boxes_filt, pred_phrases
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser("Grounding DINO example", add_help=True)
+    parser.add_argument("--config_file", "-c", type=str, required=True, help="path to config file")
+    parser.add_argument(
+        "--checkpoint_path", "-p", type=str, required=True, help="path to checkpoint file"
+    )
+    parser.add_argument("--image_path", "-i", type=str, required=True, help="path to image file")
+    parser.add_argument("--text_prompt", "-t", type=str, required=True, help="text prompt")
+    parser.add_argument(
+        "--output_dir", "-o", type=str, default="outputs", required=True, help="output directory"
+    )
+
+    parser.add_argument("--box_threshold", type=float, default=0.3, help="box threshold")
+    parser.add_argument("--text_threshold", type=float, default=0.25, help="text threshold")
+    parser.add_argument("--token_spans", type=str, default=None, help=
+                        "The positions of start and end positions of phrases of interest. \
+                        For example, a caption is 'a cat and a dog', \
+                        if you would like to detect 'cat', the token_spans should be '[[[2, 5]], ]', since 'a cat and a dog'[2:5] is 'cat'. \
+                        if you would like to detect 'a cat', the token_spans should be '[[[0, 1], [2, 5]], ]', since 'a cat and a dog'[0:1] is 'a', and 'a cat and a dog'[2:5] is 'cat'. \
+                        ")
+
+    parser.add_argument("--cpu-only", action="store_true", help="running on cpu only!, default=False")
+    args = parser.parse_args()
+
+    # cfg
+    config_file = args.config_file  # change the path of the model config file
+    checkpoint_path = args.checkpoint_path  # change the path of the model
+    image_path = args.image_path
+    text_prompt = args.text_prompt
+    output_dir = args.output_dir
+    box_threshold = args.box_threshold
+    text_threshold = args.text_threshold
+    token_spans = args.token_spans
+
+    # make dir
+    os.makedirs(output_dir, exist_ok=True)
+    # load image
+    image_pil, image = load_image(image_path)
+    # load model
+    model = load_model(config_file, checkpoint_path, cpu_only=args.cpu_only)
+
+    # visualize raw image
+    image_pil.save(os.path.join(output_dir, "raw_image.jpg"))
+
+    # set the text_threshold to None if token_spans is set.
+    if token_spans is not None:
+        text_threshold = None
+        print("Using token_spans. Set the text_threshold to None.")
+
+
+    # run model
+    boxes_filt, pred_phrases = get_grounding_output(
+        model, image, text_prompt, box_threshold, text_threshold, cpu_only=args.cpu_only, token_spans=eval(f"{token_spans}")
+    )
+
+    # visualize pred
+    size = image_pil.size
+    pred_dict = {
+        "boxes": boxes_filt,
+        "size": [size[1], size[0]],  # H,W
+        "labels": pred_phrases,
+    }
+    # import ipdb; ipdb.set_trace()
+    image_with_box = plot_boxes_to_image(image_pil, pred_dict)[0]
+    image_with_box.save(os.path.join(output_dir, "pred.jpg"))

tool/demo/test_ap_on_coco.py

@@ -0,0 +1,233 @@
+import argparse
+import os
+import sys
+import time
+
+import numpy as np
+import torch
+import torch.nn as nn
+from torch.utils.data import DataLoader, DistributedSampler
+
+from groundingdino.models import build_model
+import groundingdino.datasets.transforms as T
+from groundingdino.util import box_ops, get_tokenlizer
+from groundingdino.util.misc import clean_state_dict, collate_fn
+from groundingdino.util.slconfig import SLConfig
+
+# from torchvision.datasets import CocoDetection
+import torchvision
+
+from groundingdino.util.vl_utils import build_captions_and_token_span, create_positive_map_from_span
+from groundingdino.datasets.cocogrounding_eval import CocoGroundingEvaluator
+
+
+def load_model(model_config_path: str, model_checkpoint_path: str, device: str = "cuda"):
+    args = SLConfig.fromfile(model_config_path)
+    args.device = device
+    model = build_model(args)
+    checkpoint = torch.load(model_checkpoint_path, map_location="cpu")
+    model.load_state_dict(clean_state_dict(checkpoint["model"]), strict=False)
+    model.eval()
+    return model
+
+
+class CocoDetection(torchvision.datasets.CocoDetection):
+    def __init__(self, img_folder, ann_file, transforms):
+        super().__init__(img_folder, ann_file)
+        self._transforms = transforms
+
+    def __getitem__(self, idx):
+        img, target = super().__getitem__(idx)  # target: list
+
+        # import ipdb; ipdb.set_trace()
+
+        w, h = img.size
+        boxes = [obj["bbox"] for obj in target]
+        boxes = torch.as_tensor(boxes, dtype=torch.float32).reshape(-1, 4)
+        boxes[:, 2:] += boxes[:, :2]  # xywh -> xyxy
+        boxes[:, 0::2].clamp_(min=0, max=w)
+        boxes[:, 1::2].clamp_(min=0, max=h)
+        # filt invalid boxes/masks/keypoints
+        keep = (boxes[:, 3] > boxes[:, 1]) & (boxes[:, 2] > boxes[:, 0])
+        boxes = boxes[keep]
+
+        target_new = {}
+        image_id = self.ids[idx]
+        target_new["image_id"] = image_id
+        target_new["boxes"] = boxes
+        target_new["orig_size"] = torch.as_tensor([int(h), int(w)])
+
+        if self._transforms is not None:
+            img, target = self._transforms(img, target_new)
+
+        return img, target
+
+
+class PostProcessCocoGrounding(nn.Module):
+    """ This module converts the model's output into the format expected by the coco api"""
+
+    def __init__(self, num_select=300, coco_api=None, tokenlizer=None) -> None:
+        super().__init__()
+        self.num_select = num_select
+
+        assert coco_api is not None
+        category_dict = coco_api.dataset['categories']
+        cat_list = [item['name'] for item in category_dict]
+        captions, cat2tokenspan = build_captions_and_token_span(cat_list, True)
+        tokenspanlist = [cat2tokenspan[cat] for cat in cat_list]
+        positive_map = create_positive_map_from_span(
+            tokenlizer(captions), tokenspanlist)  # 80, 256. normed
+
+        id_map = {0: 1, 1: 2, 2: 3, 3: 4, 4: 5, 5: 6, 6: 7, 7: 8, 8: 9, 9: 10, 10: 11, 11: 13, 12: 14, 13: 15, 14: 16, 15: 17, 16: 18, 17: 19, 18: 20, 19: 21, 20: 22, 21: 23, 22: 24, 23: 25, 24: 27, 25: 28, 26: 31, 27: 32, 28: 33, 29: 34, 30: 35, 31: 36, 32: 37, 33: 38, 34: 39, 35: 40, 36: 41, 37: 42, 38: 43, 39: 44, 40: 46,
+                  41: 47, 42: 48, 43: 49, 44: 50, 45: 51, 46: 52, 47: 53, 48: 54, 49: 55, 50: 56, 51: 57, 52: 58, 53: 59, 54: 60, 55: 61, 56: 62, 57: 63, 58: 64, 59: 65, 60: 67, 61: 70, 62: 72, 63: 73, 64: 74, 65: 75, 66: 76, 67: 77, 68: 78, 69: 79, 70: 80, 71: 81, 72: 82, 73: 84, 74: 85, 75: 86, 76: 87, 77: 88, 78: 89, 79: 90}
+
+        # build a mapping from label_id to pos_map
+        new_pos_map = torch.zeros((91, 256))
+        for k, v in id_map.items():
+            new_pos_map[v] = positive_map[k]
+        self.positive_map = new_pos_map
+
+    @torch.no_grad()
+    def forward(self, outputs, target_sizes, not_to_xyxy=False):
+        """ Perform the computation
+        Parameters:
+            outputs: raw outputs of the model
+            target_sizes: tensor of dimension [batch_size x 2] containing the size of each images of the batch
+                          For evaluation, this must be the original image size (before any data augmentation)
+                          For visualization, this should be the image size after data augment, but before padding
+        """
+        num_select = self.num_select
+        out_logits, out_bbox = outputs['pred_logits'], outputs['pred_boxes']
+
+        # pos map to logit
+        prob_to_token = out_logits.sigmoid()  # bs, 100, 256
+        pos_maps = self.positive_map.to(prob_to_token.device)
+        # (bs, 100, 256) @ (91, 256).T -> (bs, 100, 91)
+        prob_to_label = prob_to_token @ pos_maps.T
+
+        # if os.environ.get('IPDB_SHILONG_DEBUG', None) == 'INFO':
+        #     import ipdb; ipdb.set_trace()
+
+        assert len(out_logits) == len(target_sizes)
+        assert target_sizes.shape[1] == 2
+
+        prob = prob_to_label
+        topk_values, topk_indexes = torch.topk(
+            prob.view(out_logits.shape[0], -1), num_select, dim=1)
+        scores = topk_values
+        topk_boxes = topk_indexes // prob.shape[2]
+        labels = topk_indexes % prob.shape[2]
+
+        if not_to_xyxy:
+            boxes = out_bbox
+        else:
+            boxes = box_ops.box_cxcywh_to_xyxy(out_bbox)
+
+        boxes = torch.gather(
+            boxes, 1, topk_boxes.unsqueeze(-1).repeat(1, 1, 4))
+
+        # and from relative [0, 1] to absolute [0, height] coordinates
+        img_h, img_w = target_sizes.unbind(1)
+        scale_fct = torch.stack([img_w, img_h, img_w, img_h], dim=1)
+        boxes = boxes * scale_fct[:, None, :]
+
+        results = [{'scores': s, 'labels': l, 'boxes': b}
+                   for s, l, b in zip(scores, labels, boxes)]
+
+        return results
+
+
+def main(args):
+    # config
+    cfg = SLConfig.fromfile(args.config_file)
+
+    # build model
+    model = load_model(args.config_file, args.checkpoint_path)
+    model = model.to(args.device)
+    model = model.eval()
+
+    # build dataloader
+    transform = T.Compose(
+        [
+            T.RandomResize([800], max_size=1333),
+            T.ToTensor(),
+            T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
+        ]
+    )
+    dataset = CocoDetection(
+        args.image_dir, args.anno_path, transforms=transform)
+    data_loader = DataLoader(
+        dataset, batch_size=1, shuffle=False, num_workers=args.num_workers, collate_fn=collate_fn)
+
+    # build post processor
+    tokenlizer = get_tokenlizer.get_tokenlizer(cfg.text_encoder_type)
+    postprocessor = PostProcessCocoGrounding(
+        coco_api=dataset.coco, tokenlizer=tokenlizer)
+
+    # build evaluator
+    evaluator = CocoGroundingEvaluator(
+        dataset.coco, iou_types=("bbox",), useCats=True)
+
+    # build captions
+    category_dict = dataset.coco.dataset['categories']
+    cat_list = [item['name'] for item in category_dict]
+    caption = " . ".join(cat_list) + ' .'
+    print("Input text prompt:", caption)
+
+    # run inference
+    start = time.time()
+    for i, (images, targets) in enumerate(data_loader):
+        # get images and captions
+        images = images.tensors.to(args.device)
+        bs = images.shape[0]
+        input_captions = [caption] * bs
+
+        # feed to the model
+        outputs = model(images, captions=input_captions)
+
+        orig_target_sizes = torch.stack(
+            [t["orig_size"] for t in targets], dim=0).to(images.device)
+        results = postprocessor(outputs, orig_target_sizes)
+        cocogrounding_res = {
+            target["image_id"]: output for target, output in zip(targets, results)}
+        evaluator.update(cocogrounding_res)
+
+        if (i+1) % 30 == 0:
+            used_time = time.time() - start
+            eta = len(data_loader) / (i+1e-5) * used_time - used_time
+            print(
+                f"processed {i}/{len(data_loader)} images. time: {used_time:.2f}s, ETA: {eta:.2f}s")
+
+    evaluator.synchronize_between_processes()
+    evaluator.accumulate()
+    evaluator.summarize()
+
+    print("Final results:", evaluator.coco_eval["bbox"].stats.tolist())
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        "Grounding DINO eval on COCO", add_help=True)
+    # load model
+    parser.add_argument("--config_file", "-c", type=str,
+                        required=True, help="path to config file")
+    parser.add_argument(
+        "--checkpoint_path", "-p", type=str, required=True, help="path to checkpoint file"
+    )
+    parser.add_argument("--device", type=str, default="cuda",
+                        help="running device (default: cuda)")
+
+    # post processing
+    parser.add_argument("--num_select", type=int, default=300,
+                        help="number of topk to select")
+
+    # coco info
+    parser.add_argument("--anno_path", type=str,
+                        required=True, help="coco root")
+    parser.add_argument("--image_dir", type=str,
+                        required=True, help="coco image dir")
+    parser.add_argument("--num_workers", type=int, default=4,
+                        help="number of workers for dataloader")
+    args = parser.parse_args()
+
+    main(args)

tool/detector.py

@@ -0,0 +1,93 @@
+import torch
+import numpy as np
+import cv2
+import PIL
+
+from groundingdino.models import build_model as build_grounding_dino
+from groundingdino.util.slconfig import SLConfig
+from groundingdino.util.utils import clean_state_dict
+from groundingdino.util.inference import annotate, load_image, predict
+import groundingdino.datasets.transforms as T
+
+from torchvision.ops import box_convert
+
+class Detector:
+    def __init__(self, device):
+        config_file = "src/groundingdino/groundingdino/config/GroundingDINO_SwinT_OGC.py"
+        grounding_dino_ckpt = './ckpt/groundingdino_swint_ogc.pth'
+        args = SLConfig.fromfile(config_file) 
+        args.device = device
+        self.deivce = device
+        self.gd = build_grounding_dino(args)
+
+        checkpoint = torch.load(grounding_dino_ckpt, map_location='cpu')
+        log = self.gd.load_state_dict(clean_state_dict(checkpoint['model']), strict=False)
+        print("Model loaded from {} \n => {}".format(grounding_dino_ckpt, log))
+        self.gd.eval()
+    
+    def image_transform_grounding(self, init_image):
+        transform = T.Compose([
+            T.RandomResize([800], max_size=1333),
+            T.ToTensor(),
+            T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
+        ])
+        image, _ = transform(init_image, None) # 3, h, w
+        return init_image, image
+
+    def image_transform_grounding_for_vis(self, init_image):
+        transform = T.Compose([
+            T.RandomResize([800], max_size=1333),
+        ])
+        image, _ = transform(init_image, None) # 3, h, w
+        return image
+
+    def transfer_boxes_format(self, boxes, height, width):
+        boxes = boxes * torch.Tensor([width, height, width, height])
+        boxes = box_convert(boxes=boxes, in_fmt="cxcywh", out_fmt="xyxy")
+
+        transfered_boxes = []
+        for i in range(len(boxes)):
+            box = boxes[i]
+            transfered_box = [[int(box[0]), int(box[1])], [int(box[2]), int(box[3])]]
+            transfered_boxes.append(transfered_box)
+        
+        transfered_boxes = np.array(transfered_boxes)
+        return transfered_boxes
+        
+    @torch.no_grad()
+    def run_grounding(self, origin_frame, grounding_caption, box_threshold, text_threshold):
+        '''
+            return:
+                annotated_frame:nd.array
+                transfered_boxes: nd.array [N, 4]: [[x0, y0], [x1, y1]]
+        '''
+        height, width, _ = origin_frame.shape
+        img_pil = PIL.Image.fromarray(origin_frame)
+        re_width, re_height = img_pil.size
+        _, image_tensor = self.image_transform_grounding(img_pil)
+        # img_pil = self.image_transform_grounding_for_vis(img_pil)
+
+        # run grounidng
+        boxes, logits, phrases = predict(self.gd, image_tensor, grounding_caption, box_threshold, text_threshold, device=self.deivce)
+        annotated_frame = annotate(image_source=np.asarray(img_pil), boxes=boxes, logits=logits, phrases=phrases)[:, :, ::-1]
+        annotated_frame = cv2.resize(annotated_frame, (width, height), interpolation=cv2.INTER_LINEAR)
+        
+        # transfer boxes to sam-format 
+        transfered_boxes = self.transfer_boxes_format(boxes, re_height, re_width)
+        return annotated_frame, transfered_boxes
+
+if __name__ == "__main__":
+    detector = Detector("cuda")
+    origin_frame = cv2.imread('./debug/point.png')
+    origin_frame = cv2.cvtColor(origin_frame, cv2.COLOR_BGR2RGB)
+    grounding_caption = "swan.water"
+    box_threshold = 0.25
+    text_threshold = 0.25
+
+    annotated_frame, boxes = detector.run_grounding(origin_frame, grounding_caption, box_threshold, text_threshold)
+    cv2.imwrite('./debug/x.png', annotated_frame)
+
+    for i in range(len(boxes)):
+        bbox = boxes[i]
+        origin_frame = cv2.rectangle(origin_frame, bbox[0], bbox[1], (0, 0, 255))
+    cv2.imwrite('./debug/bbox_frame.png', origin_frame)

tool/environment.yaml

@@ -0,0 +1,248 @@
+name: dino
+channels:
+  - pytorch
+  - nvidia
+  - conda-forge
+  - defaults
+dependencies:
+  - addict=2.4.0=pyhd8ed1ab_2
+  - aiohttp=3.8.5=py39ha55989b_0
+  - aiosignal=1.3.1=pyhd8ed1ab_0
+  - asttokens=2.0.5=pyhd3eb1b0_0
+  - async-timeout=4.0.3=pyhd8ed1ab_0
+  - attrs=23.1.0=pyh71513ae_1
+  - aws-c-auth=0.7.0=h6f3c987_2
+  - aws-c-cal=0.6.0=h6ba3258_0
+  - aws-c-common=0.8.23=hcfcfb64_0
+  - aws-c-compression=0.2.17=h420beca_1
+  - aws-c-event-stream=0.3.1=had47b81_1
+  - aws-c-http=0.7.11=h72ba615_0
+  - aws-c-io=0.13.28=ha35c040_0
+  - aws-c-mqtt=0.8.14=h4941efa_2
+  - aws-c-s3=0.3.13=he04eaa7_2
+  - aws-c-sdkutils=0.1.11=h420beca_1
+  - aws-checksums=0.1.16=h420beca_1
+  - aws-crt-cpp=0.20.3=h247a981_4
+  - aws-sdk-cpp=1.10.57=h1a0519f_17
+  - backcall=0.2.0=pyhd3eb1b0_0
+  - blas=2.118=mkl
+  - blas-devel=3.9.0=18_win64_mkl
+  - brotli=1.0.9=hcfcfb64_9
+  - brotli-bin=1.0.9=hcfcfb64_9
+  - brotli-python=1.0.9=py39h99910a6_9
+  - bzip2=1.0.8=h8ffe710_4
+  - c-ares=1.19.1=hcfcfb64_0
+  - ca-certificates=2023.08.22=haa95532_0
+  - certifi=2023.7.22=py39haa95532_0
+  - charset-normalizer=3.2.0=pyhd8ed1ab_0
+  - click=8.1.7=win_pyh7428d3b_0
+  - colorama=0.4.6=pyhd8ed1ab_0
+  - comm=0.1.2=py39haa95532_0
+  - contourpy=1.1.1=py39h1f6ef14_1
+  - cuda-cccl=12.2.140=0
+  - cuda-cudart=11.8.89=0
+  - cuda-cudart-dev=11.8.89=0
+  - cuda-cupti=11.8.87=0
+  - cuda-libraries=11.8.0=0
+  - cuda-libraries-dev=11.8.0=0
+  - cuda-nvrtc=11.8.89=0
+  - cuda-nvrtc-dev=11.8.89=0
+  - cuda-nvtx=11.8.86=0
+  - cuda-profiler-api=12.2.140=0
+  - cuda-runtime=11.8.0=0
+  - cycler=0.11.0=pyhd8ed1ab_0
+  - cython=3.0.0=py39h2bbff1b_0
+  - dataclasses=0.8=pyhc8e2a94_3
+  - datasets=2.14.5=pyhd8ed1ab_0
+  - debugpy=1.6.7=py39hd77b12b_0
+  - decorator=5.1.1=pyhd3eb1b0_0
+  - dill=0.3.7=pyhd8ed1ab_0
+  - exceptiongroup=1.0.4=py39haa95532_0
+  - executing=0.8.3=pyhd3eb1b0_0
+  - filelock=3.12.4=pyhd8ed1ab_0
+  - fonttools=4.42.1=py39ha55989b_0
+  - freeglut=3.2.2=h63175ca_2
+  - freetype=2.12.1=hdaf720e_2
+  - frozenlist=1.4.0=py39ha55989b_1
+  - fsspec=2023.6.0=pyh1a96a4e_0
+  - gettext=0.21.1=h5728263_0
+  - glib=2.78.0=h12be248_0
+  - glib-tools=2.78.0=h12be248_0
+  - gst-plugins-base=1.22.6=h001b923_1
+  - gstreamer=1.22.6=hb4038d2_1
+  - huggingface_hub=0.17.3=pyhd8ed1ab_0
+  - icu=70.1=h0e60522_0
+  - idna=3.4=pyhd8ed1ab_0
+  - importlib-metadata=6.8.0=pyha770c72_0
+  - importlib-resources=6.1.0=pyhd8ed1ab_0
+  - importlib_metadata=6.8.0=hd8ed1ab_0
+  - importlib_resources=6.1.0=pyhd8ed1ab_0
+  - intel-openmp=2023.2.0=h57928b3_49503
+  - ipykernel=6.25.0=py39h9909e9c_0
+  - ipython=8.15.0=py39haa95532_0
+  - jasper=2.0.33=hc2e4405_1
+  - jedi=0.18.1=py39haa95532_1
+  - jinja2=3.1.2=pyhd8ed1ab_1
+  - joblib=1.3.2=pyhd8ed1ab_0
+  - jpeg=9e=hcfcfb64_3
+  - jupyter_client=8.1.0=py39haa95532_0
+  - jupyter_core=5.3.0=py39haa95532_0
+  - kiwisolver=1.4.5=py39h1f6ef14_1
+  - krb5=1.20.1=heb0366b_0
+  - lcms2=2.14=h90d422f_0
+  - lerc=4.0.0=h63175ca_0
+  - libabseil=20230125.3=cxx17_h63175ca_0
+  - libarrow=12.0.1=h12e5d06_5_cpu
+  - libblas=3.9.0=18_win64_mkl
+  - libbrotlicommon=1.0.9=hcfcfb64_9
+  - libbrotlidec=1.0.9=hcfcfb64_9
+  - libbrotlienc=1.0.9=hcfcfb64_9
+  - libcblas=3.9.0=18_win64_mkl
+  - libclang=15.0.7=default_h77d9078_3
+  - libclang13=15.0.7=default_h77d9078_3
+  - libcrc32c=1.1.2=h0e60522_0
+  - libcublas=11.11.3.6=0
+  - libcublas-dev=11.11.3.6=0
+  - libcufft=10.9.0.58=0
+  - libcufft-dev=10.9.0.58=0
+  - libcurand=10.3.3.141=0
+  - libcurand-dev=10.3.3.141=0
+  - libcurl=8.1.2=h68f0423_0
+  - libcusolver=11.4.1.48=0
+  - libcusolver-dev=11.4.1.48=0
+  - libcusparse=11.7.5.86=0
+  - libcusparse-dev=11.7.5.86=0
+  - libdeflate=1.14=hcfcfb64_0
+  - libevent=2.1.12=h3671451_1
+  - libffi=3.4.2=h8ffe710_5
+  - libglib=2.78.0=he8f3873_0
+  - libgoogle-cloud=2.12.0=h00b2bdc_1
+  - libgrpc=1.54.3=ha177ca7_0
+  - libhwloc=2.9.3=default_haede6df_1009
+  - libiconv=1.17=h8ffe710_0
+  - liblapack=3.9.0=18_win64_mkl
+  - liblapacke=3.9.0=18_win64_mkl
+  - libnpp=11.8.0.86=0
+  - libnpp-dev=11.8.0.86=0
+  - libnvjpeg=11.9.0.86=0
+  - libnvjpeg-dev=11.9.0.86=0
+  - libogg=1.3.4=h8ffe710_1
+  - libopencv=4.5.3=py39h488c12c_8
+  - libpng=1.6.39=h19919ed_0
+  - libprotobuf=3.21.12=h12be248_2
+  - libsodium=1.0.18=h62dcd97_0
+  - libsqlite=3.43.0=hcfcfb64_0
+  - libssh2=1.11.0=h7dfc565_0
+  - libthrift=0.18.1=h06f6336_2
+  - libtiff=4.4.0=hc4f729c_5
+  - libutf8proc=2.8.0=h82a8f57_0
+  - libuv=1.44.2=hcfcfb64_1
+  - libvorbis=1.3.7=h0e60522_0
+  - libwebp-base=1.3.2=hcfcfb64_0
+  - libxcb=1.13=hcd874cb_1004
+  - libxml2=2.11.5=hc3477c8_1
+  - libzlib=1.2.13=hcfcfb64_5
+  - lz4-c=1.9.4=hcfcfb64_0
+  - m2w64-gcc-libgfortran=5.3.0=6
+  - m2w64-gcc-libs=5.3.0=7
+  - m2w64-gcc-libs-core=5.3.0=7
+  - m2w64-gmp=6.1.0=2
+  - m2w64-libwinpthread-git=5.0.0.4634.697f757=2
+  - markupsafe=2.1.3=py39ha55989b_1
+  - matplotlib-base=3.8.0=py39hf19769e_1
+  - matplotlib-inline=0.1.6=py39haa95532_0
+  - mkl=2022.1.0=h6a75c08_874
+  - mkl-devel=2022.1.0=h57928b3_875
+  - mkl-include=2022.1.0=h6a75c08_874
+  - mpmath=1.3.0=pyhd8ed1ab_0
+  - msys2-conda-epoch=20160418=1
+  - multidict=6.0.4=py39ha55989b_0
+  - multiprocess=0.70.15=py39ha55989b_1
+  - munkres=1.1.4=pyh9f0ad1d_0
+  - nest-asyncio=1.5.6=py39haa95532_0
+  - networkx=3.1=pyhd8ed1ab_0
+  - numpy=1.26.0=py39hddb5d58_0
+  - opencv=4.5.3=py39hcbf5309_8
+  - openjpeg=2.5.0=hc9384bd_1
+  - openssl=3.1.3=hcfcfb64_0
+  - orc=1.9.0=hada7b9e_1
+  - packaging=23.1=pyhd8ed1ab_0
+  - pandas=2.1.1=py39h32e6231_0
+  - parso=0.8.3=pyhd3eb1b0_0
+  - pcre2=10.40=h17e33f8_0
+  - pickleshare=0.7.5=pyhd3eb1b0_1003
+  - pillow=9.2.0=py39h595c93f_3
+  - pip=23.2.1=pyhd8ed1ab_0
+  - platformdirs=3.10.0=pyhd8ed1ab_0
+  - prompt-toolkit=3.0.36=py39haa95532_0
+  - psutil=5.9.0=py39h2bbff1b_0
+  - pthread-stubs=0.4=hcd874cb_1001
+  - pthreads-win32=2.9.1=hfa6e2cd_3
+  - pure_eval=0.2.2=pyhd3eb1b0_0
+  - py-opencv=4.5.3=py39h00e5391_8
+  - pyarrow=12.0.1=py39hca4e8af_5_cpu
+  - pycocotools=2.0.6=py39hc266a54_1
+  - pygments=2.15.1=py39haa95532_1
+  - pyparsing=3.1.1=pyhd8ed1ab_0
+  - pysocks=1.7.1=pyh0701188_6
+  - python=3.9.18=h4de0772_0_cpython
+  - python-dateutil=2.8.2=pyhd8ed1ab_0
+  - python-tzdata=2023.3=pyhd8ed1ab_0
+  - python-xxhash=3.3.0=py39ha55989b_1
+  - python_abi=3.9=4_cp39
+  - pytorch=2.0.1=py3.9_cuda11.8_cudnn8_0
+  - pytorch-cuda=11.8=h24eeafa_5
+  - pytorch-mutex=1.0=cuda
+  - pytz=2023.3.post1=pyhd8ed1ab_0
+  - pywin32=305=py39h2bbff1b_0
+  - pyyaml=6.0.1=py39ha55989b_1
+  - pyzmq=25.1.0=py39hd77b12b_0
+  - qt-main=5.15.8=h720456b_6
+  - re2=2023.03.02=hd4eee63_0
+  - regex=2023.8.8=py39ha55989b_1
+  - requests=2.31.0=pyhd8ed1ab_0
+  - sacremoses=0.0.53=pyhd8ed1ab_0
+  - safetensors=0.3.3=py39hf21820d_1
+  - setuptools=68.2.2=pyhd8ed1ab_0
+  - six=1.16.0=pyh6c4a22f_0
+  - snappy=1.1.10=hfb803bf_0
+  - stack_data=0.2.0=pyhd3eb1b0_0
+  - sympy=1.12=pyh04b8f61_3
+  - tbb=2021.10.0=h91493d7_1
+  - timm=0.9.7=pyhd8ed1ab_0
+  - tk=8.6.13=hcfcfb64_0
+  - tokenizers=0.13.3=py39hca44cb7_0
+  - tomli=2.0.1=pyhd8ed1ab_0
+  - tornado=6.3.2=py39h2bbff1b_0
+  - tqdm=4.66.1=pyhd8ed1ab_0
+  - traitlets=5.7.1=py39haa95532_0
+  - transformers=4.33.2=pyhd8ed1ab_0
+  - typing-extensions=4.8.0=hd8ed1ab_0
+  - typing_extensions=4.8.0=pyha770c72_0
+  - tzdata=2023c=h71feb2d_0
+  - ucrt=10.0.22621.0=h57928b3_0
+  - unicodedata2=15.0.0=py39ha55989b_1
+  - urllib3=2.0.5=pyhd8ed1ab_0
+  - vc=14.3=h64f974e_17
+  - vc14_runtime=14.36.32532=hdcecf7f_17
+  - vs2015_runtime=14.36.32532=h05e6639_17
+  - wcwidth=0.2.5=pyhd3eb1b0_0
+  - wheel=0.41.2=pyhd8ed1ab_0
+  - win_inet_pton=1.1.0=pyhd8ed1ab_6
+  - xorg-libxau=1.0.11=hcd874cb_0
+  - xorg-libxdmcp=1.1.3=hcd874cb_0
+  - xxhash=0.8.2=hcfcfb64_0
+  - xz=5.2.6=h8d14728_0
+  - yaml=0.2.5=h8ffe710_2
+  - yapf=0.40.1=pyhd8ed1ab_0
+  - yarl=1.9.2=py39ha55989b_0
+  - zeromq=4.3.4=hd77b12b_0
+  - zipp=3.17.0=pyhd8ed1ab_0
+  - zlib=1.2.13=hcfcfb64_5
+  - zstd=1.5.5=h12be248_0
+  - pip:
+      - opencv-python==4.8.0.76
+      - supervision==0.6.0
+      - torchaudio==2.0.2
+      - torchvision==0.15.2
+prefix: C:\Users\Makoto\miniconda3\envs\dino

tool/groundingdino.egg-info/PKG-INFO

@@ -0,0 +1,209 @@
+Metadata-Version: 2.1
+Name: groundingdino
+Version: 0.1.0
+Summary: open-set object detector
+Home-page: https://github.com/IDEA-Research/GroundingDINO
+Author: International Digital Economy Academy, Shilong Liu
+License:                                  Apache License
+                                   Version 2.0, January 2004
+                                http://www.apache.org/licenses/
+        
+           TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+        
+           1. Definitions.
+        
+              "License" shall mean the terms and conditions for use, reproduction,
+              and distribution as defined by Sections 1 through 9 of this document.
+        
+              "Licensor" shall mean the copyright owner or entity authorized by
+              the copyright owner that is granting the License.
+        
+              "Legal Entity" shall mean the union of the acting entity and all
+              other entities that control, are controlled by, or are under common
+              control with that entity. For the purposes of this definition,
+              "control" means (i) the power, direct or indirect, to cause the
+              direction or management of such entity, whether by contract or
+              otherwise, or (ii) ownership of fifty percent (50%) or more of the
+              outstanding shares, or (iii) beneficial ownership of such entity.
+        
+              "You" (or "Your") shall mean an individual or Legal Entity
+              exercising permissions granted by this License.
+        
+              "Source" form shall mean the preferred form for making modifications,
+              including but not limited to software source code, documentation
+              source, and configuration files.
+        
+              "Object" form shall mean any form resulting from mechanical
+              transformation or translation of a Source form, including but
+              not limited to compiled object code, generated documentation,
+              and conversions to other media types.
+        
+              "Work" shall mean the work of authorship, whether in Source or
+              Object form, made available under the License, as indicated by a
+              copyright notice that is included in or attached to the work
+              (an example is provided in the Appendix below).
+        
+              "Derivative Works" shall mean any work, whether in Source or Object
+              form, that is based on (or derived from) the Work and for which the
+              editorial revisions, annotations, elaborations, or other modifications
+              represent, as a whole, an original work of authorship. For the purposes
+              of this License, Derivative Works shall not include works that remain
+              separable from, or merely link (or bind by name) to the interfaces of,
+              the Work and Derivative Works thereof.
+        
+              "Contribution" shall mean any work of authorship, including
+              the original version of the Work and any modifications or additions
+              to that Work or Derivative Works thereof, that is intentionally
+              submitted to Licensor for inclusion in the Work by the copyright owner
+              or by an individual or Legal Entity authorized to submit on behalf of
+              the copyright owner. For the purposes of this definition, "submitted"
+              means any form of electronic, verbal, or written communication sent
+              to the Licensor or its representatives, including but not limited to
+              communication on electronic mailing lists, source code control systems,
+              and issue tracking systems that are managed by, or on behalf of, the
+              Licensor for the purpose of discussing and improving the Work, but
+              excluding communication that is conspicuously marked or otherwise
+              designated in writing by the copyright owner as "Not a Contribution."
+        
+              "Contributor" shall mean Licensor and any individual or Legal Entity
+              on behalf of whom a Contribution has been received by Licensor and
+              subsequently incorporated within the Work.
+        
+           2. Grant of Copyright License. Subject to the terms and conditions of
+              this License, each Contributor hereby grants to You a perpetual,
+              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+              copyright license to reproduce, prepare Derivative Works of,
+              publicly display, publicly perform, sublicense, and distribute the
+              Work and such Derivative Works in Source or Object form.
+        
+           3. Grant of Patent License. Subject to the terms and conditions of
+              this License, each Contributor hereby grants to You a perpetual,
+              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+              (except as stated in this section) patent license to make, have made,
+              use, offer to sell, sell, import, and otherwise transfer the Work,
+              where such license applies only to those patent claims licensable
+              by such Contributor that are necessarily infringed by their
+              Contribution(s) alone or by combination of their Contribution(s)
+              with the Work to which such Contribution(s) was submitted. If You
+              institute patent litigation against any entity (including a
+              cross-claim or counterclaim in a lawsuit) alleging that the Work
+              or a Contribution incorporated within the Work constitutes direct
+              or contributory patent infringement, then any patent licenses
+              granted to You under this License for that Work shall terminate
+              as of the date such litigation is filed.
+        
+           4. Redistribution. You may reproduce and distribute copies of the
+              Work or Derivative Works thereof in any medium, with or without
+              modifications, and in Source or Object form, provided that You
+              meet the following conditions:
+        
+              (a) You must give any other recipients of the Work or
+                  Derivative Works a copy of this License; and
+        
+              (b) You must cause any modified files to carry prominent notices
+                  stating that You changed the files; and
+        
+              (c) You must retain, in the Source form of any Derivative Works
+                  that You distribute, all copyright, patent, trademark, and
+                  attribution notices from the Source form of the Work,
+                  excluding those notices that do not pertain to any part of
+                  the Derivative Works; and
+        
+              (d) If the Work includes a "NOTICE" text file as part of its
+                  distribution, then any Derivative Works that You distribute must
+                  include a readable copy of the attribution notices contained
+                  within such NOTICE file, excluding those notices that do not
+                  pertain to any part of the Derivative Works, in at least one
+                  of the following places: within a NOTICE text file distributed
+                  as part of the Derivative Works; within the Source form or
+                  documentation, if provided along with the Derivative Works; or,
+                  within a display generated by the Derivative Works, if and
+                  wherever such third-party notices normally appear. The contents
+                  of the NOTICE file are for informational purposes only and
+                  do not modify the License. You may add Your own attribution
+                  notices within Derivative Works that You distribute, alongside
+                  or as an addendum to the NOTICE text from the Work, provided
+                  that such additional attribution notices cannot be construed
+                  as modifying the License.
+        
+              You may add Your own copyright statement to Your modifications and
+              may provide additional or different license terms and conditions
+              for use, reproduction, or distribution of Your modifications, or
+              for any such Derivative Works as a whole, provided Your use,
+              reproduction, and distribution of the Work otherwise complies with
+              the conditions stated in this License.
+        
+           5. Submission of Contributions. Unless You explicitly state otherwise,
+              any Contribution intentionally submitted for inclusion in the Work
+              by You to the Licensor shall be under the terms and conditions of
+              this License, without any additional terms or conditions.
+              Notwithstanding the above, nothing herein shall supersede or modify
+              the terms of any separate license agreement you may have executed
+              with Licensor regarding such Contributions.
+        
+           6. Trademarks. This License does not grant permission to use the trade
+              names, trademarks, service marks, or product names of the Licensor,
+              except as required for reasonable and customary use in describing the
+              origin of the Work and reproducing the content of the NOTICE file.
+        
+           7. Disclaimer of Warranty. Unless required by applicable law or
+              agreed to in writing, Licensor provides the Work (and each
+              Contributor provides its Contributions) on an "AS IS" BASIS,
+              WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+              implied, including, without limitation, any warranties or conditions
+              of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+              PARTICULAR PURPOSE. You are solely responsible for determining the
+              appropriateness of using or redistributing the Work and assume any
+              risks associated with Your exercise of permissions under this License.
+        
+           8. Limitation of Liability. In no event and under no legal theory,
+              whether in tort (including negligence), contract, or otherwise,
+              unless required by applicable law (such as deliberate and grossly
+              negligent acts) or agreed to in writing, shall any Contributor be
+              liable to You for damages, including any direct, indirect, special,
+              incidental, or consequential damages of any character arising as a
+              result of this License or out of the use or inability to use the
+              Work (including but not limited to damages for loss of goodwill,
+              work stoppage, computer failure or malfunction, or any and all
+              other commercial damages or losses), even if such Contributor
+              has been advised of the possibility of such damages.
+        
+           9. Accepting Warranty or Additional Liability. While redistributing
+              the Work or Derivative Works thereof, You may choose to offer,
+              and charge a fee for, acceptance of support, warranty, indemnity,
+              or other liability obligations and/or rights consistent with this
+              License. However, in accepting such obligations, You may act only
+              on Your own behalf and on Your sole responsibility, not on behalf
+              of any other Contributor, and only if You agree to indemnify,
+              defend, and hold each Contributor harmless for any liability
+              incurred by, or claims asserted against, such Contributor by reason
+              of your accepting any such warranty or additional liability.
+        
+           END OF TERMS AND CONDITIONS
+        
+           APPENDIX: How to apply the Apache License to your work.
+        
+              To apply the Apache License to your work, attach the following
+              boilerplate notice, with the fields enclosed by brackets "[]"
+              replaced with your own identifying information. (Don't include
+              the brackets!)  The text should be enclosed in the appropriate
+              comment syntax for the file format. We also recommend that a
+              file or class name and description of purpose be included on the
+              same "printed page" as the copyright notice for easier
+              identification within third-party archives.
+        
+           Copyright 2023 - present, IDEA Research.
+        
+           Licensed under the Apache License, Version 2.0 (the "License");
+           you may not use this file except in compliance with the License.
+           You may obtain a copy of the License at
+        
+               http://www.apache.org/licenses/LICENSE-2.0
+        
+           Unless required by applicable law or agreed to in writing, software
+           distributed under the License is distributed on an "AS IS" BASIS,
+           WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+           See the License for the specific language governing permissions and
+           limitations under the License.
+        
+License-File: LICENSE

tool/groundingdino.egg-info/SOURCES.txt

@@ -0,0 +1,42 @@
+LICENSE
+README.md
+setup.py
+groundingdino/__init__.py
+groundingdino/version.py
+groundingdino.egg-info/PKG-INFO
+groundingdino.egg-info/SOURCES.txt
+groundingdino.egg-info/dependency_links.txt
+groundingdino.egg-info/requires.txt
+groundingdino.egg-info/top_level.txt
+groundingdino/config/GroundingDINO_SwinB_cfg.py
+groundingdino/config/GroundingDINO_SwinT_OGC.py
+groundingdino/config/__init__.py
+groundingdino/datasets/__init__.py
+groundingdino/datasets/cocogrounding_eval.py
+groundingdino/datasets/transforms.py
+groundingdino/models/__init__.py
+groundingdino/models/registry.py
+groundingdino/models/GroundingDINO/__init__.py
+groundingdino/models/GroundingDINO/bertwarper.py
+groundingdino/models/GroundingDINO/fuse_modules.py
+groundingdino/models/GroundingDINO/groundingdino.py
+groundingdino/models/GroundingDINO/ms_deform_attn.py
+groundingdino/models/GroundingDINO/transformer.py
+groundingdino/models/GroundingDINO/transformer_vanilla.py
+groundingdino/models/GroundingDINO/utils.py
+groundingdino/models/GroundingDINO/backbone/__init__.py
+groundingdino/models/GroundingDINO/backbone/backbone.py
+groundingdino/models/GroundingDINO/backbone/position_encoding.py
+groundingdino/models/GroundingDINO/backbone/swin_transformer.py
+groundingdino/util/__init__.py
+groundingdino/util/box_ops.py
+groundingdino/util/get_tokenlizer.py
+groundingdino/util/inference.py
+groundingdino/util/logger.py
+groundingdino/util/misc.py
+groundingdino/util/slconfig.py
+groundingdino/util/slio.py
+groundingdino/util/time_counter.py
+groundingdino/util/utils.py
+groundingdino/util/visualizer.py
+groundingdino/util/vl_utils.py

tool/groundingdino.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

tool/groundingdino.egg-info/requires.txt

@@ -0,0 +1,10 @@
+torch
+torchvision
+transformers
+addict
+yapf
+timm
+numpy
+opencv-python
+supervision==0.6.0
+pycocotools

tool/groundingdino.egg-info/top_level.txt

@@ -0,0 +1 @@
+groundingdino

tool/groundingdino/config/GroundingDINO_SwinB_cfg.py

@@ -0,0 +1,43 @@
+batch_size = 1
+modelname = "groundingdino"
+backbone = "swin_B_384_22k"
+position_embedding = "sine"
+pe_temperatureH = 20
+pe_temperatureW = 20
+return_interm_indices = [1, 2, 3]
+backbone_freeze_keywords = None
+enc_layers = 6
+dec_layers = 6
+pre_norm = False
+dim_feedforward = 2048
+hidden_dim = 256
+dropout = 0.0
+nheads = 8
+num_queries = 900
+query_dim = 4
+num_patterns = 0
+num_feature_levels = 4
+enc_n_points = 4
+dec_n_points = 4
+two_stage_type = "standard"
+two_stage_bbox_embed_share = False
+two_stage_class_embed_share = False
+transformer_activation = "relu"
+dec_pred_bbox_embed_share = True
+dn_box_noise_scale = 1.0
+dn_label_noise_ratio = 0.5
+dn_label_coef = 1.0
+dn_bbox_coef = 1.0
+embed_init_tgt = True
+dn_labelbook_size = 2000
+max_text_len = 256
+text_encoder_type = "bert-base-uncased"
+use_text_enhancer = True
+use_fusion_layer = True
+use_checkpoint = True
+use_transformer_ckpt = True
+use_text_cross_attention = True
+text_dropout = 0.0
+fusion_dropout = 0.0
+fusion_droppath = 0.1
+sub_sentence_present = True

tool/groundingdino/config/GroundingDINO_SwinT_OGC.py

@@ -0,0 +1,43 @@
+batch_size = 1
+modelname = "groundingdino"
+backbone = "swin_T_224_1k"
+position_embedding = "sine"
+pe_temperatureH = 20
+pe_temperatureW = 20
+return_interm_indices = [1, 2, 3]
+backbone_freeze_keywords = None
+enc_layers = 6
+dec_layers = 6
+pre_norm = False
+dim_feedforward = 2048
+hidden_dim = 256
+dropout = 0.0
+nheads = 8
+num_queries = 900
+query_dim = 4
+num_patterns = 0
+num_feature_levels = 4
+enc_n_points = 4
+dec_n_points = 4
+two_stage_type = "standard"
+two_stage_bbox_embed_share = False
+two_stage_class_embed_share = False
+transformer_activation = "relu"
+dec_pred_bbox_embed_share = True
+dn_box_noise_scale = 1.0
+dn_label_noise_ratio = 0.5
+dn_label_coef = 1.0
+dn_bbox_coef = 1.0
+embed_init_tgt = True
+dn_labelbook_size = 2000
+max_text_len = 256
+text_encoder_type = "bert-base-uncased"
+use_text_enhancer = True
+use_fusion_layer = True
+use_checkpoint = True
+use_transformer_ckpt = True
+use_text_cross_attention = True
+text_dropout = 0.0
+fusion_dropout = 0.0
+fusion_droppath = 0.1
+sub_sentence_present = True

tool/groundingdino/datasets/cocogrounding_eval.py

@@ -0,0 +1,269 @@
+# ------------------------------------------------------------------------
+# Grounding DINO. Midified by Shilong Liu.
+# url: https://github.com/IDEA-Research/GroundingDINO
+# Copyright (c) 2023 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Copyright (c) Aishwarya Kamath & Nicolas Carion. Licensed under the Apache License 2.0. All Rights Reserved
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
+"""
+COCO evaluator that works in distributed mode.
+
+Mostly copy-paste from https://github.com/pytorch/vision/blob/edfd5a7/references/detection/coco_eval.py
+The difference is that there is less copy-pasting from pycocotools
+in the end of the file, as python3 can suppress prints with contextlib
+"""
+import contextlib
+import copy
+import os
+
+import numpy as np
+import pycocotools.mask as mask_util
+import torch
+from pycocotools.coco import COCO
+from pycocotools.cocoeval import COCOeval
+
+from groundingdino.util.misc import all_gather
+
+
+class CocoGroundingEvaluator(object):
+    def __init__(self, coco_gt, iou_types, useCats=True):
+        assert isinstance(iou_types, (list, tuple))
+        coco_gt = copy.deepcopy(coco_gt)
+        self.coco_gt = coco_gt
+
+        self.iou_types = iou_types
+        self.coco_eval = {}
+        for iou_type in iou_types:
+            self.coco_eval[iou_type] = COCOeval(coco_gt, iouType=iou_type)
+            self.coco_eval[iou_type].useCats = useCats
+
+        self.img_ids = []
+        self.eval_imgs = {k: [] for k in iou_types}
+        self.useCats = useCats
+
+    def update(self, predictions):
+        img_ids = list(np.unique(list(predictions.keys())))
+        self.img_ids.extend(img_ids)
+
+        for iou_type in self.iou_types:
+            results = self.prepare(predictions, iou_type)
+
+            # suppress pycocotools prints
+            with open(os.devnull, "w") as devnull:
+                with contextlib.redirect_stdout(devnull):
+                    coco_dt = COCO.loadRes(self.coco_gt, results) if results else COCO()
+
+            coco_eval = self.coco_eval[iou_type]
+
+            coco_eval.cocoDt = coco_dt
+            coco_eval.params.imgIds = list(img_ids)
+            coco_eval.params.useCats = self.useCats
+            img_ids, eval_imgs = evaluate(coco_eval)
+
+            self.eval_imgs[iou_type].append(eval_imgs)
+
+    def synchronize_between_processes(self):
+        for iou_type in self.iou_types:
+            self.eval_imgs[iou_type] = np.concatenate(self.eval_imgs[iou_type], 2)
+            create_common_coco_eval(self.coco_eval[iou_type], self.img_ids, self.eval_imgs[iou_type])
+
+    def accumulate(self):
+        for coco_eval in self.coco_eval.values():
+            coco_eval.accumulate()
+
+    def summarize(self):
+        for iou_type, coco_eval in self.coco_eval.items():
+            print("IoU metric: {}".format(iou_type))
+            coco_eval.summarize()
+
+    def prepare(self, predictions, iou_type):
+        if iou_type == "bbox":
+            return self.prepare_for_coco_detection(predictions)
+        elif iou_type == "segm":
+            return self.prepare_for_coco_segmentation(predictions)
+        elif iou_type == "keypoints":
+            return self.prepare_for_coco_keypoint(predictions)
+        else:
+            raise ValueError("Unknown iou type {}".format(iou_type))
+
+    def prepare_for_coco_detection(self, predictions):
+        coco_results = []
+        for original_id, prediction in predictions.items():
+            if len(prediction) == 0:
+                continue
+
+            boxes = prediction["boxes"]
+            boxes = convert_to_xywh(boxes).tolist()
+            scores = prediction["scores"].tolist()
+            labels = prediction["labels"].tolist()
+
+            coco_results.extend(
+                [
+                    {
+                        "image_id": original_id,
+                        "category_id": labels[k],
+                        "bbox": box,
+                        "score": scores[k],
+                    }
+                    for k, box in enumerate(boxes)
+                ]
+            )
+        return coco_results
+
+    def prepare_for_coco_segmentation(self, predictions):
+        coco_results = []
+        for original_id, prediction in predictions.items():
+            if len(prediction) == 0:
+                continue
+
+            scores = prediction["scores"]
+            labels = prediction["labels"]
+            masks = prediction["masks"]
+
+            masks = masks > 0.5
+
+            scores = prediction["scores"].tolist()
+            labels = prediction["labels"].tolist()
+
+            rles = [
+                mask_util.encode(np.array(mask[0, :, :, np.newaxis], dtype=np.uint8, order="F"))[0] 
+                for mask in masks
+            ]
+            for rle in rles:
+                rle["counts"] = rle["counts"].decode("utf-8")
+
+            coco_results.extend(
+                [
+                    {
+                        "image_id": original_id,
+                        "category_id": labels[k],
+                        "segmentation": rle,
+                        "score": scores[k],
+                    }
+                    for k, rle in enumerate(rles)
+                ]
+            )
+        return coco_results
+
+    def prepare_for_coco_keypoint(self, predictions):
+        coco_results = []
+        for original_id, prediction in predictions.items():
+            if len(prediction) == 0:
+                continue
+
+            boxes = prediction["boxes"]
+            boxes = convert_to_xywh(boxes).tolist()
+            scores = prediction["scores"].tolist()
+            labels = prediction["labels"].tolist()
+            keypoints = prediction["keypoints"]
+            keypoints = keypoints.flatten(start_dim=1).tolist()
+
+            coco_results.extend(
+                [
+                    {
+                        "image_id": original_id,
+                        "category_id": labels[k],
+                        "keypoints": keypoint,
+                        "score": scores[k],
+                    }
+                    for k, keypoint in enumerate(keypoints)
+                ]
+            )
+        return coco_results
+
+
+def convert_to_xywh(boxes):
+    xmin, ymin, xmax, ymax = boxes.unbind(1)
+    return torch.stack((xmin, ymin, xmax - xmin, ymax - ymin), dim=1)
+
+
+def merge(img_ids, eval_imgs):
+    all_img_ids = all_gather(img_ids)
+    all_eval_imgs = all_gather(eval_imgs)
+
+    merged_img_ids = []
+    for p in all_img_ids:
+        merged_img_ids.extend(p)
+
+    merged_eval_imgs = []
+    for p in all_eval_imgs:
+        merged_eval_imgs.append(p)
+
+    merged_img_ids = np.array(merged_img_ids)
+    merged_eval_imgs = np.concatenate(merged_eval_imgs, 2)
+
+    # keep only unique (and in sorted order) images
+    merged_img_ids, idx = np.unique(merged_img_ids, return_index=True)
+    merged_eval_imgs = merged_eval_imgs[..., idx]
+
+    return merged_img_ids, merged_eval_imgs
+
+
+def create_common_coco_eval(coco_eval, img_ids, eval_imgs):
+    img_ids, eval_imgs = merge(img_ids, eval_imgs)
+    img_ids = list(img_ids)
+    eval_imgs = list(eval_imgs.flatten())
+
+    coco_eval.evalImgs = eval_imgs
+    coco_eval.params.imgIds = img_ids
+    coco_eval._paramsEval = copy.deepcopy(coco_eval.params)
+
+
+#################################################################
+# From pycocotools, just removed the prints and fixed
+# a Python3 bug about unicode not defined
+#################################################################
+
+
+def evaluate(self):
+    """
+    Run per image evaluation on given images and store results (a list of dict) in self.evalImgs
+    :return: None
+    """
+    # tic = time.time()
+    # print('Running per image evaluation...')
+    p = self.params
+    # add backward compatibility if useSegm is specified in params
+    if p.useSegm is not None:
+        p.iouType = "segm" if p.useSegm == 1 else "bbox"
+        print("useSegm (deprecated) is not None. Running {} evaluation".format(p.iouType))
+    # print('Evaluate annotation type *{}*'.format(p.iouType))
+    p.imgIds = list(np.unique(p.imgIds))
+    if p.useCats:
+        p.catIds = list(np.unique(p.catIds))
+    p.maxDets = sorted(p.maxDets)
+    self.params = p
+
+    self._prepare()
+    # loop through images, area range, max detection number
+    catIds = p.catIds if p.useCats else [-1]
+
+    if p.iouType == "segm" or p.iouType == "bbox":
+        computeIoU = self.computeIoU
+    elif p.iouType == "keypoints":
+        computeIoU = self.computeOks
+    self.ious = {
+        (imgId, catId): computeIoU(imgId, catId) 
+        for imgId in p.imgIds 
+        for catId in catIds}
+
+    evaluateImg = self.evaluateImg
+    maxDet = p.maxDets[-1]
+    evalImgs = [
+        evaluateImg(imgId, catId, areaRng, maxDet) 
+        for catId in catIds 
+        for areaRng in p.areaRng 
+        for imgId in p.imgIds
+    ]
+    # this is NOT in the pycocotools code, but could be done outside
+    evalImgs = np.asarray(evalImgs).reshape(len(catIds), len(p.areaRng), len(p.imgIds))
+    self._paramsEval = copy.deepcopy(self.params)
+    # toc = time.time()
+    # print('DONE (t={:0.2f}s).'.format(toc-tic))
+    return p.imgIds, evalImgs
+
+
+#################################################################
+# end of straight copy from pycocotools, just removing the prints
+#################################################################

tool/groundingdino/datasets/transforms.py

@@ -0,0 +1,311 @@
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
+"""
+Transforms and data augmentation for both image + bbox.
+"""
+import os
+import random
+
+import PIL
+import torch
+import torchvision.transforms as T
+import torchvision.transforms.functional as F
+
+from groundingdino.util.box_ops import box_xyxy_to_cxcywh
+from groundingdino.util.misc import interpolate
+
+
+def crop(image, target, region):
+    cropped_image = F.crop(image, *region)
+
+    target = target.copy()
+    i, j, h, w = region
+
+    # should we do something wrt the original size?
+    target["size"] = torch.tensor([h, w])
+
+    fields = ["labels", "area", "iscrowd", "positive_map"]
+
+    if "boxes" in target:
+        boxes = target["boxes"]
+        max_size = torch.as_tensor([w, h], dtype=torch.float32)
+        cropped_boxes = boxes - torch.as_tensor([j, i, j, i])
+        cropped_boxes = torch.min(cropped_boxes.reshape(-1, 2, 2), max_size)
+        cropped_boxes = cropped_boxes.clamp(min=0)
+        area = (cropped_boxes[:, 1, :] - cropped_boxes[:, 0, :]).prod(dim=1)
+        target["boxes"] = cropped_boxes.reshape(-1, 4)
+        target["area"] = area
+        fields.append("boxes")
+
+    if "masks" in target:
+        # FIXME should we update the area here if there are no boxes?
+        target["masks"] = target["masks"][:, i : i + h, j : j + w]
+        fields.append("masks")
+
+    # remove elements for which the boxes or masks that have zero area
+    if "boxes" in target or "masks" in target:
+        # favor boxes selection when defining which elements to keep
+        # this is compatible with previous implementation
+        if "boxes" in target:
+            cropped_boxes = target["boxes"].reshape(-1, 2, 2)
+            keep = torch.all(cropped_boxes[:, 1, :] > cropped_boxes[:, 0, :], dim=1)
+        else:
+            keep = target["masks"].flatten(1).any(1)
+
+        for field in fields:
+            if field in target:
+                target[field] = target[field][keep]
+
+    if os.environ.get("IPDB_SHILONG_DEBUG", None) == "INFO":
+        # for debug and visualization only.
+        if "strings_positive" in target:
+            target["strings_positive"] = [
+                _i for _i, _j in zip(target["strings_positive"], keep) if _j
+            ]
+
+    return cropped_image, target
+
+
+def hflip(image, target):
+    flipped_image = F.hflip(image)
+
+    w, h = image.size
+
+    target = target.copy()
+    if "boxes" in target:
+        boxes = target["boxes"]
+        boxes = boxes[:, [2, 1, 0, 3]] * torch.as_tensor([-1, 1, -1, 1]) + torch.as_tensor(
+            [w, 0, w, 0]
+        )
+        target["boxes"] = boxes
+
+    if "masks" in target:
+        target["masks"] = target["masks"].flip(-1)
+
+    return flipped_image, target
+
+
+def resize(image, target, size, max_size=None):
+    # size can be min_size (scalar) or (w, h) tuple
+
+    def get_size_with_aspect_ratio(image_size, size, max_size=None):
+        w, h = image_size
+        if max_size is not None:
+            min_original_size = float(min((w, h)))
+            max_original_size = float(max((w, h)))
+            if max_original_size / min_original_size * size > max_size:
+                size = int(round(max_size * min_original_size / max_original_size))
+
+        if (w <= h and w == size) or (h <= w and h == size):
+            return (h, w)
+
+        if w < h:
+            ow = size
+            oh = int(size * h / w)
+        else:
+            oh = size
+            ow = int(size * w / h)
+
+        return (oh, ow)
+
+    def get_size(image_size, size, max_size=None):
+        if isinstance(size, (list, tuple)):
+            return size[::-1]
+        else:
+            return get_size_with_aspect_ratio(image_size, size, max_size)
+
+    size = get_size(image.size, size, max_size)
+    rescaled_image = F.resize(image, size)
+
+    if target is None:
+        return rescaled_image, None
+
+    ratios = tuple(float(s) / float(s_orig) for s, s_orig in zip(rescaled_image.size, image.size))
+    ratio_width, ratio_height = ratios
+
+    target = target.copy()
+    if "boxes" in target:
+        boxes = target["boxes"]
+        scaled_boxes = boxes * torch.as_tensor(
+            [ratio_width, ratio_height, ratio_width, ratio_height]
+        )
+        target["boxes"] = scaled_boxes
+
+    if "area" in target:
+        area = target["area"]
+        scaled_area = area * (ratio_width * ratio_height)
+        target["area"] = scaled_area
+
+    h, w = size
+    target["size"] = torch.tensor([h, w])
+
+    if "masks" in target:
+        target["masks"] = (
+            interpolate(target["masks"][:, None].float(), size, mode="nearest")[:, 0] > 0.5
+        )
+
+    return rescaled_image, target
+
+
+def pad(image, target, padding):
+    # assumes that we only pad on the bottom right corners
+    padded_image = F.pad(image, (0, 0, padding[0], padding[1]))
+    if target is None:
+        return padded_image, None
+    target = target.copy()
+    # should we do something wrt the original size?
+    target["size"] = torch.tensor(padded_image.size[::-1])
+    if "masks" in target:
+        target["masks"] = torch.nn.functional.pad(target["masks"], (0, padding[0], 0, padding[1]))
+    return padded_image, target
+
+
+class ResizeDebug(object):
+    def __init__(self, size):
+        self.size = size
+
+    def __call__(self, img, target):
+        return resize(img, target, self.size)
+
+
+class RandomCrop(object):
+    def __init__(self, size):
+        self.size = size
+
+    def __call__(self, img, target):
+        region = T.RandomCrop.get_params(img, self.size)
+        return crop(img, target, region)
+
+
+class RandomSizeCrop(object):
+    def __init__(self, min_size: int, max_size: int, respect_boxes: bool = False):
+        # respect_boxes:    True to keep all boxes
+        #                   False to tolerence box filter
+        self.min_size = min_size
+        self.max_size = max_size
+        self.respect_boxes = respect_boxes
+
+    def __call__(self, img: PIL.Image.Image, target: dict):
+        init_boxes = len(target["boxes"])
+        max_patience = 10
+        for i in range(max_patience):
+            w = random.randint(self.min_size, min(img.width, self.max_size))
+            h = random.randint(self.min_size, min(img.height, self.max_size))
+            region = T.RandomCrop.get_params(img, [h, w])
+            result_img, result_target = crop(img, target, region)
+            if (
+                not self.respect_boxes
+                or len(result_target["boxes"]) == init_boxes
+                or i == max_patience - 1
+            ):
+                return result_img, result_target
+        return result_img, result_target
+
+
+class CenterCrop(object):
+    def __init__(self, size):
+        self.size = size
+
+    def __call__(self, img, target):
+        image_width, image_height = img.size
+        crop_height, crop_width = self.size
+        crop_top = int(round((image_height - crop_height) / 2.0))
+        crop_left = int(round((image_width - crop_width) / 2.0))
+        return crop(img, target, (crop_top, crop_left, crop_height, crop_width))
+
+
+class RandomHorizontalFlip(object):
+    def __init__(self, p=0.5):
+        self.p = p
+
+    def __call__(self, img, target):
+        if random.random() < self.p:
+            return hflip(img, target)
+        return img, target
+
+
+class RandomResize(object):
+    def __init__(self, sizes, max_size=None):
+        assert isinstance(sizes, (list, tuple))
+        self.sizes = sizes
+        self.max_size = max_size
+
+    def __call__(self, img, target=None):
+        size = random.choice(self.sizes)
+        return resize(img, target, size, self.max_size)
+
+
+class RandomPad(object):
+    def __init__(self, max_pad):
+        self.max_pad = max_pad
+
+    def __call__(self, img, target):
+        pad_x = random.randint(0, self.max_pad)
+        pad_y = random.randint(0, self.max_pad)
+        return pad(img, target, (pad_x, pad_y))
+
+
+class RandomSelect(object):
+    """
+    Randomly selects between transforms1 and transforms2,
+    with probability p for transforms1 and (1 - p) for transforms2
+    """
+
+    def __init__(self, transforms1, transforms2, p=0.5):
+        self.transforms1 = transforms1
+        self.transforms2 = transforms2
+        self.p = p
+
+    def __call__(self, img, target):
+        if random.random() < self.p:
+            return self.transforms1(img, target)
+        return self.transforms2(img, target)
+
+
+class ToTensor(object):
+    def __call__(self, img, target):
+        return F.to_tensor(img), target
+
+
+class RandomErasing(object):
+    def __init__(self, *args, **kwargs):
+        self.eraser = T.RandomErasing(*args, **kwargs)
+
+    def __call__(self, img, target):
+        return self.eraser(img), target
+
+
+class Normalize(object):
+    def __init__(self, mean, std):
+        self.mean = mean
+        self.std = std
+
+    def __call__(self, image, target=None):
+        image = F.normalize(image, mean=self.mean, std=self.std)
+        if target is None:
+            return image, None
+        target = target.copy()
+        h, w = image.shape[-2:]
+        if "boxes" in target:
+            boxes = target["boxes"]
+            boxes = box_xyxy_to_cxcywh(boxes)
+            boxes = boxes / torch.tensor([w, h, w, h], dtype=torch.float32)
+            target["boxes"] = boxes
+        return image, target
+
+
+class Compose(object):
+    def __init__(self, transforms):
+        self.transforms = transforms
+
+    def __call__(self, image, target):
+        for t in self.transforms:
+            image, target = t(image, target)
+        return image, target
+
+    def __repr__(self):
+        format_string = self.__class__.__name__ + "("
+        for t in self.transforms:
+            format_string += "\n"
+            format_string += "    {0}".format(t)
+        format_string += "\n)"
+        return format_string

tool/groundingdino/models/GroundingDINO/__init__.py

@@ -0,0 +1,15 @@
+# ------------------------------------------------------------------------
+# Grounding DINO
+# url: https://github.com/IDEA-Research/GroundingDINO
+# Copyright (c) 2023 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Conditional DETR
+# Copyright (c) 2021 Microsoft. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Copied from DETR (https://github.com/facebookresearch/detr)
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+# ------------------------------------------------------------------------
+
+from .groundingdino import build_groundingdino

tool/groundingdino/models/GroundingDINO/backbone/__init__.py

@@ -0,0 +1 @@
+from .backbone import build_backbone

tool/groundingdino/models/GroundingDINO/backbone/backbone.py

@@ -0,0 +1,221 @@
+# ------------------------------------------------------------------------
+# Grounding DINO
+# url: https://github.com/IDEA-Research/GroundingDINO
+# Copyright (c) 2023 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Conditional DETR
+# Copyright (c) 2021 Microsoft. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Copied from DETR (https://github.com/facebookresearch/detr)
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+# ------------------------------------------------------------------------
+
+"""
+Backbone modules.
+"""
+
+from typing import Dict, List
+
+import torch
+import torch.nn.functional as F
+import torchvision
+from torch import nn
+from torchvision.models._utils import IntermediateLayerGetter
+
+from groundingdino.util.misc import NestedTensor, clean_state_dict, is_main_process
+
+from .position_encoding import build_position_encoding
+from .swin_transformer import build_swin_transformer
+
+
+class FrozenBatchNorm2d(torch.nn.Module):
+    """
+    BatchNorm2d where the batch statistics and the affine parameters are fixed.
+
+    Copy-paste from torchvision.misc.ops with added eps before rqsrt,
+    without which any other models than torchvision.models.resnet[18,34,50,101]
+    produce nans.
+    """
+
+    def __init__(self, n):
+        super(FrozenBatchNorm2d, self).__init__()
+        self.register_buffer("weight", torch.ones(n))
+        self.register_buffer("bias", torch.zeros(n))
+        self.register_buffer("running_mean", torch.zeros(n))
+        self.register_buffer("running_var", torch.ones(n))
+
+    def _load_from_state_dict(
+        self, state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs
+    ):
+        num_batches_tracked_key = prefix + "num_batches_tracked"
+        if num_batches_tracked_key in state_dict:
+            del state_dict[num_batches_tracked_key]
+
+        super(FrozenBatchNorm2d, self)._load_from_state_dict(
+            state_dict, prefix, local_metadata, strict, missing_keys, unexpected_keys, error_msgs
+        )
+
+    def forward(self, x):
+        # move reshapes to the beginning
+        # to make it fuser-friendly
+        w = self.weight.reshape(1, -1, 1, 1)
+        b = self.bias.reshape(1, -1, 1, 1)
+        rv = self.running_var.reshape(1, -1, 1, 1)
+        rm = self.running_mean.reshape(1, -1, 1, 1)
+        eps = 1e-5
+        scale = w * (rv + eps).rsqrt()
+        bias = b - rm * scale
+        return x * scale + bias
+
+
+class BackboneBase(nn.Module):
+    def __init__(
+        self,
+        backbone: nn.Module,
+        train_backbone: bool,
+        num_channels: int,
+        return_interm_indices: list,
+    ):
+        super().__init__()
+        for name, parameter in backbone.named_parameters():
+            if (
+                not train_backbone
+                or "layer2" not in name
+                and "layer3" not in name
+                and "layer4" not in name
+            ):
+                parameter.requires_grad_(False)
+
+        return_layers = {}
+        for idx, layer_index in enumerate(return_interm_indices):
+            return_layers.update(
+                {"layer{}".format(5 - len(return_interm_indices) + idx): "{}".format(layer_index)}
+            )
+
+        # if len:
+        #     if use_stage1_feature:
+        #         return_layers = {"layer1": "0", "layer2": "1", "layer3": "2", "layer4": "3"}
+        #     else:
+        #         return_layers = {"layer2": "0", "layer3": "1", "layer4": "2"}
+        # else:
+        #     return_layers = {'layer4': "0"}
+        self.body = IntermediateLayerGetter(backbone, return_layers=return_layers)
+        self.num_channels = num_channels
+
+    def forward(self, tensor_list: NestedTensor):
+        xs = self.body(tensor_list.tensors)
+        out: Dict[str, NestedTensor] = {}
+        for name, x in xs.items():
+            m = tensor_list.mask
+            assert m is not None
+            mask = F.interpolate(m[None].float(), size=x.shape[-2:]).to(torch.bool)[0]
+            out[name] = NestedTensor(x, mask)
+        # import ipdb; ipdb.set_trace()
+        return out
+
+
+class Backbone(BackboneBase):
+    """ResNet backbone with frozen BatchNorm."""
+
+    def __init__(
+        self,
+        name: str,
+        train_backbone: bool,
+        dilation: bool,
+        return_interm_indices: list,
+        batch_norm=FrozenBatchNorm2d,
+    ):
+        if name in ["resnet18", "resnet34", "resnet50", "resnet101"]:
+            backbone = getattr(torchvision.models, name)(
+                replace_stride_with_dilation=[False, False, dilation],
+                pretrained=is_main_process(),
+                norm_layer=batch_norm,
+            )
+        else:
+            raise NotImplementedError("Why you can get here with name {}".format(name))
+        # num_channels = 512 if name in ('resnet18', 'resnet34') else 2048
+        assert name not in ("resnet18", "resnet34"), "Only resnet50 and resnet101 are available."
+        assert return_interm_indices in [[0, 1, 2, 3], [1, 2, 3], [3]]
+        num_channels_all = [256, 512, 1024, 2048]
+        num_channels = num_channels_all[4 - len(return_interm_indices) :]
+        super().__init__(backbone, train_backbone, num_channels, return_interm_indices)
+
+
+class Joiner(nn.Sequential):
+    def __init__(self, backbone, position_embedding):
+        super().__init__(backbone, position_embedding)
+
+    def forward(self, tensor_list: NestedTensor):
+        xs = self[0](tensor_list)
+        out: List[NestedTensor] = []
+        pos = []
+        for name, x in xs.items():
+            out.append(x)
+            # position encoding
+            pos.append(self[1](x).to(x.tensors.dtype))
+
+        return out, pos
+
+
+def build_backbone(args):
+    """
+    Useful args:
+        - backbone: backbone name
+        - lr_backbone:
+        - dilation
+        - return_interm_indices: available: [0,1,2,3], [1,2,3], [3]
+        - backbone_freeze_keywords:
+        - use_checkpoint: for swin only for now
+
+    """
+    position_embedding = build_position_encoding(args)
+    train_backbone = True
+    if not train_backbone:
+        raise ValueError("Please set lr_backbone > 0")
+    return_interm_indices = args.return_interm_indices
+    assert return_interm_indices in [[0, 1, 2, 3], [1, 2, 3], [3]]
+    args.backbone_freeze_keywords
+    use_checkpoint = getattr(args, "use_checkpoint", False)
+
+    if args.backbone in ["resnet50", "resnet101"]:
+        backbone = Backbone(
+            args.backbone,
+            train_backbone,
+            args.dilation,
+            return_interm_indices,
+            batch_norm=FrozenBatchNorm2d,
+        )
+        bb_num_channels = backbone.num_channels
+    elif args.backbone in [
+        "swin_T_224_1k",
+        "swin_B_224_22k",
+        "swin_B_384_22k",
+        "swin_L_224_22k",
+        "swin_L_384_22k",
+    ]:
+        pretrain_img_size = int(args.backbone.split("_")[-2])
+        backbone = build_swin_transformer(
+            args.backbone,
+            pretrain_img_size=pretrain_img_size,
+            out_indices=tuple(return_interm_indices),
+            dilation=False,
+            use_checkpoint=use_checkpoint,
+        )
+
+        bb_num_channels = backbone.num_features[4 - len(return_interm_indices) :]
+    else:
+        raise NotImplementedError("Unknown backbone {}".format(args.backbone))
+
+    assert len(bb_num_channels) == len(
+        return_interm_indices
+    ), f"len(bb_num_channels) {len(bb_num_channels)} != len(return_interm_indices) {len(return_interm_indices)}"
+
+    model = Joiner(backbone, position_embedding)
+    model.num_channels = bb_num_channels
+    assert isinstance(
+        bb_num_channels, List
+    ), "bb_num_channels is expected to be a List but {}".format(type(bb_num_channels))
+    # import ipdb; ipdb.set_trace()
+    return model

tool/groundingdino/models/GroundingDINO/backbone/position_encoding.py

@@ -0,0 +1,186 @@
+# ------------------------------------------------------------------------
+# Grounding DINO
+# url: https://github.com/IDEA-Research/GroundingDINO
+# Copyright (c) 2023 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# DINO
+# Copyright (c) 2022 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Conditional DETR
+# Copyright (c) 2021 Microsoft. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# Copied from DETR (https://github.com/facebookresearch/detr)
+# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+# ------------------------------------------------------------------------
+
+"""
+Various positional encodings for the transformer.
+"""
+import math
+
+import torch
+from torch import nn
+
+from groundingdino.util.misc import NestedTensor
+
+
+class PositionEmbeddingSine(nn.Module):
+    """
+    This is a more standard version of the position embedding, very similar to the one
+    used by the Attention is all you need paper, generalized to work on images.
+    """
+
+    def __init__(self, num_pos_feats=64, temperature=10000, normalize=False, scale=None):
+        super().__init__()
+        self.num_pos_feats = num_pos_feats
+        self.temperature = temperature
+        self.normalize = normalize
+        if scale is not None and normalize is False:
+            raise ValueError("normalize should be True if scale is passed")
+        if scale is None:
+            scale = 2 * math.pi
+        self.scale = scale
+
+    def forward(self, tensor_list: NestedTensor):
+        x = tensor_list.tensors
+        mask = tensor_list.mask
+        assert mask is not None
+        not_mask = ~mask
+        y_embed = not_mask.cumsum(1, dtype=torch.float32)
+        x_embed = not_mask.cumsum(2, dtype=torch.float32)
+        if self.normalize:
+            eps = 1e-6
+            # if os.environ.get("SHILONG_AMP", None) == '1':
+            #     eps = 1e-4
+            # else:
+            #     eps = 1e-6
+            y_embed = y_embed / (y_embed[:, -1:, :] + eps) * self.scale
+            x_embed = x_embed / (x_embed[:, :, -1:] + eps) * self.scale
+
+        dim_t = torch.arange(self.num_pos_feats, dtype=torch.float32, device=x.device)
+        dim_t = self.temperature ** (2 * (dim_t // 2) / self.num_pos_feats)
+
+        pos_x = x_embed[:, :, :, None] / dim_t
+        pos_y = y_embed[:, :, :, None] / dim_t
+        pos_x = torch.stack(
+            (pos_x[:, :, :, 0::2].sin(), pos_x[:, :, :, 1::2].cos()), dim=4
+        ).flatten(3)
+        pos_y = torch.stack(
+            (pos_y[:, :, :, 0::2].sin(), pos_y[:, :, :, 1::2].cos()), dim=4
+        ).flatten(3)
+        pos = torch.cat((pos_y, pos_x), dim=3).permute(0, 3, 1, 2)
+        return pos
+
+
+class PositionEmbeddingSineHW(nn.Module):
+    """
+    This is a more standard version of the position embedding, very similar to the one
+    used by the Attention is all you need paper, generalized to work on images.
+    """
+
+    def __init__(
+        self, num_pos_feats=64, temperatureH=10000, temperatureW=10000, normalize=False, scale=None
+    ):
+        super().__init__()
+        self.num_pos_feats = num_pos_feats
+        self.temperatureH = temperatureH
+        self.temperatureW = temperatureW
+        self.normalize = normalize
+        if scale is not None and normalize is False:
+            raise ValueError("normalize should be True if scale is passed")
+        if scale is None:
+            scale = 2 * math.pi
+        self.scale = scale
+
+    def forward(self, tensor_list: NestedTensor):
+        x = tensor_list.tensors
+        mask = tensor_list.mask
+        assert mask is not None
+        not_mask = ~mask
+        y_embed = not_mask.cumsum(1, dtype=torch.float32)
+        x_embed = not_mask.cumsum(2, dtype=torch.float32)
+
+        # import ipdb; ipdb.set_trace()
+
+        if self.normalize:
+            eps = 1e-6
+            y_embed = y_embed / (y_embed[:, -1:, :] + eps) * self.scale
+            x_embed = x_embed / (x_embed[:, :, -1:] + eps) * self.scale
+
+        dim_tx = torch.arange(self.num_pos_feats, dtype=torch.float32, device=x.device)
+        dim_tx = self.temperatureW ** (2 * (torch.div(dim_tx, 2, rounding_mode='floor')) / self.num_pos_feats)
+        pos_x = x_embed[:, :, :, None] / dim_tx
+
+        dim_ty = torch.arange(self.num_pos_feats, dtype=torch.float32, device=x.device)
+        dim_ty = self.temperatureH ** (2 * (torch.div(dim_ty, 2, rounding_mode='floor')) / self.num_pos_feats)
+        pos_y = y_embed[:, :, :, None] / dim_ty
+
+        pos_x = torch.stack(
+            (pos_x[:, :, :, 0::2].sin(), pos_x[:, :, :, 1::2].cos()), dim=4
+        ).flatten(3)
+        pos_y = torch.stack(
+            (pos_y[:, :, :, 0::2].sin(), pos_y[:, :, :, 1::2].cos()), dim=4
+        ).flatten(3)
+        pos = torch.cat((pos_y, pos_x), dim=3).permute(0, 3, 1, 2)
+
+        # import ipdb; ipdb.set_trace()
+
+        return pos
+
+
+class PositionEmbeddingLearned(nn.Module):
+    """
+    Absolute pos embedding, learned.
+    """
+
+    def __init__(self, num_pos_feats=256):
+        super().__init__()
+        self.row_embed = nn.Embedding(50, num_pos_feats)
+        self.col_embed = nn.Embedding(50, num_pos_feats)
+        self.reset_parameters()
+
+    def reset_parameters(self):
+        nn.init.uniform_(self.row_embed.weight)
+        nn.init.uniform_(self.col_embed.weight)
+
+    def forward(self, tensor_list: NestedTensor):
+        x = tensor_list.tensors
+        h, w = x.shape[-2:]
+        i = torch.arange(w, device=x.device)
+        j = torch.arange(h, device=x.device)
+        x_emb = self.col_embed(i)
+        y_emb = self.row_embed(j)
+        pos = (
+            torch.cat(
+                [
+                    x_emb.unsqueeze(0).repeat(h, 1, 1),
+                    y_emb.unsqueeze(1).repeat(1, w, 1),
+                ],
+                dim=-1,
+            )
+            .permute(2, 0, 1)
+            .unsqueeze(0)
+            .repeat(x.shape[0], 1, 1, 1)
+        )
+        return pos
+
+
+def build_position_encoding(args):
+    N_steps = args.hidden_dim // 2
+    if args.position_embedding in ("v2", "sine"):
+        # TODO find a better way of exposing other arguments
+        position_embedding = PositionEmbeddingSineHW(
+            N_steps,
+            temperatureH=args.pe_temperatureH,
+            temperatureW=args.pe_temperatureW,
+            normalize=True,
+        )
+    elif args.position_embedding in ("v3", "learned"):
+        position_embedding = PositionEmbeddingLearned(N_steps)
+    else:
+        raise ValueError(f"not supported {args.position_embedding}")
+
+    return position_embedding

tool/groundingdino/models/GroundingDINO/backbone/swin_transformer.py

@@ -0,0 +1,802 @@
+# ------------------------------------------------------------------------
+# Grounding DINO
+# url: https://github.com/IDEA-Research/GroundingDINO
+# Copyright (c) 2023 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# ------------------------------------------------------------------------
+# DINO
+# Copyright (c) 2022 IDEA. All Rights Reserved.
+# Licensed under the Apache License, Version 2.0 [see LICENSE for details]
+# --------------------------------------------------------
+# modified from https://github.com/SwinTransformer/Swin-Transformer-Object-Detection/blob/master/mmdet/models/backbones/swin_transformer.py
+# --------------------------------------------------------
+
+import numpy as np
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import torch.utils.checkpoint as checkpoint
+from timm.models.layers import DropPath, to_2tuple, trunc_normal_
+
+from groundingdino.util.misc import NestedTensor
+
+
+class Mlp(nn.Module):
+    """Multilayer perceptron."""
+
+    def __init__(
+        self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.0
+    ):
+        super().__init__()
+        out_features = out_features or in_features
+        hidden_features = hidden_features or in_features
+        self.fc1 = nn.Linear(in_features, hidden_features)
+        self.act = act_layer()
+        self.fc2 = nn.Linear(hidden_features, out_features)
+        self.drop = nn.Dropout(drop)
+
+    def forward(self, x):
+        x = self.fc1(x)
+        x = self.act(x)
+        x = self.drop(x)
+        x = self.fc2(x)
+        x = self.drop(x)
+        return x
+
+
+def window_partition(x, window_size):
+    """
+    Args:
+        x: (B, H, W, C)
+        window_size (int): window size
+    Returns:
+        windows: (num_windows*B, window_size, window_size, C)
+    """
+    B, H, W, C = x.shape
+    x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
+    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
+    return windows
+
+
+def window_reverse(windows, window_size, H, W):
+    """
+    Args:
+        windows: (num_windows*B, window_size, window_size, C)
+        window_size (int): Window size
+        H (int): Height of image
+        W (int): Width of image
+    Returns:
+        x: (B, H, W, C)
+    """
+    B = int(windows.shape[0] / (H * W / window_size / window_size))
+    x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
+    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
+    return x
+
+
+class WindowAttention(nn.Module):
+    """Window based multi-head self attention (W-MSA) module with relative position bias.
+    It supports both of shifted and non-shifted window.
+    Args:
+        dim (int): Number of input channels.
+        window_size (tuple[int]): The height and width of the window.
+        num_heads (int): Number of attention heads.
+        qkv_bias (bool, optional):  If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set
+        attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0
+        proj_drop (float, optional): Dropout ratio of output. Default: 0.0
+    """
+
+    def __init__(
+        self,
+        dim,
+        window_size,
+        num_heads,
+        qkv_bias=True,
+        qk_scale=None,
+        attn_drop=0.0,
+        proj_drop=0.0,
+    ):
+
+        super().__init__()
+        self.dim = dim
+        self.window_size = window_size  # Wh, Ww
+        self.num_heads = num_heads
+        head_dim = dim // num_heads
+        self.scale = qk_scale or head_dim**-0.5
+
+        # define a parameter table of relative position bias
+        self.relative_position_bias_table = nn.Parameter(
+            torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads)
+        )  # 2*Wh-1 * 2*Ww-1, nH
+
+        # get pair-wise relative position index for each token inside the window
+        coords_h = torch.arange(self.window_size[0])
+        coords_w = torch.arange(self.window_size[1])
+        coords = torch.stack(torch.meshgrid([coords_h, coords_w]))  # 2, Wh, Ww
+        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
+        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 2, Wh*Ww, Wh*Ww
+        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2
+        relative_coords[:, :, 0] += self.window_size[0] - 1  # shift to start from 0
+        relative_coords[:, :, 1] += self.window_size[1] - 1
+        relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
+        relative_position_index = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
+        self.register_buffer("relative_position_index", relative_position_index)
+
+        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
+        self.attn_drop = nn.Dropout(attn_drop)
+        self.proj = nn.Linear(dim, dim)
+        self.proj_drop = nn.Dropout(proj_drop)
+
+        trunc_normal_(self.relative_position_bias_table, std=0.02)
+        self.softmax = nn.Softmax(dim=-1)
+
+    def forward(self, x, mask=None):
+        """Forward function.
+        Args:
+            x: input features with shape of (num_windows*B, N, C)
+            mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None
+        """
+        B_, N, C = x.shape
+        qkv = (
+            self.qkv(x)
+            .reshape(B_, N, 3, self.num_heads, C // self.num_heads)
+            .permute(2, 0, 3, 1, 4)
+        )
+        q, k, v = qkv[0], qkv[1], qkv[2]  # make torchscript happy (cannot use tensor as tuple)
+
+        q = q * self.scale
+        attn = q @ k.transpose(-2, -1)
+
+        relative_position_bias = self.relative_position_bias_table[
+            self.relative_position_index.view(-1)
+        ].view(
+            self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1
+        )  # Wh*Ww,Wh*Ww,nH
+        relative_position_bias = relative_position_bias.permute(
+            2, 0, 1
+        ).contiguous()  # nH, Wh*Ww, Wh*Ww
+        attn = attn + relative_position_bias.unsqueeze(0)
+
+        if mask is not None:
+            nW = mask.shape[0]
+            attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
+            attn = attn.view(-1, self.num_heads, N, N)
+            attn = self.softmax(attn)
+        else:
+            attn = self.softmax(attn)
+
+        attn = self.attn_drop(attn)
+
+        x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
+        x = self.proj(x)
+        x = self.proj_drop(x)
+        return x
+
+
+class SwinTransformerBlock(nn.Module):
+    """Swin Transformer Block.
+    Args:
+        dim (int): Number of input channels.
+        num_heads (int): Number of attention heads.
+        window_size (int): Window size.
+        shift_size (int): Shift size for SW-MSA.
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
+        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
+        drop (float, optional): Dropout rate. Default: 0.0
+        attn_drop (float, optional): Attention dropout rate. Default: 0.0
+        drop_path (float, optional): Stochastic depth rate. Default: 0.0
+        act_layer (nn.Module, optional): Activation layer. Default: nn.GELU
+        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
+    """
+
+    def __init__(
+        self,
+        dim,
+        num_heads,
+        window_size=7,
+        shift_size=0,
+        mlp_ratio=4.0,
+        qkv_bias=True,
+        qk_scale=None,
+        drop=0.0,
+        attn_drop=0.0,
+        drop_path=0.0,
+        act_layer=nn.GELU,
+        norm_layer=nn.LayerNorm,
+    ):
+        super().__init__()
+        self.dim = dim
+        self.num_heads = num_heads
+        self.window_size = window_size
+        self.shift_size = shift_size
+        self.mlp_ratio = mlp_ratio
+        assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"
+
+        self.norm1 = norm_layer(dim)
+        self.attn = WindowAttention(
+            dim,
+            window_size=to_2tuple(self.window_size),
+            num_heads=num_heads,
+            qkv_bias=qkv_bias,
+            qk_scale=qk_scale,
+            attn_drop=attn_drop,
+            proj_drop=drop,
+        )
+
+        self.drop_path = DropPath(drop_path) if drop_path > 0.0 else nn.Identity()
+        self.norm2 = norm_layer(dim)
+        mlp_hidden_dim = int(dim * mlp_ratio)
+        self.mlp = Mlp(
+            in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop
+        )
+
+        self.H = None
+        self.W = None
+
+    def forward(self, x, mask_matrix):
+        """Forward function.
+        Args:
+            x: Input feature, tensor size (B, H*W, C).
+            H, W: Spatial resolution of the input feature.
+            mask_matrix: Attention mask for cyclic shift.
+        """
+        B, L, C = x.shape
+        H, W = self.H, self.W
+        assert L == H * W, "input feature has wrong size"
+
+        shortcut = x
+        x = self.norm1(x)
+        x = x.view(B, H, W, C)
+
+        # pad feature maps to multiples of window size
+        pad_l = pad_t = 0
+        pad_r = (self.window_size - W % self.window_size) % self.window_size
+        pad_b = (self.window_size - H % self.window_size) % self.window_size
+        x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b))
+        _, Hp, Wp, _ = x.shape
+
+        # cyclic shift
+        if self.shift_size > 0:
+            shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
+            attn_mask = mask_matrix
+        else:
+            shifted_x = x
+            attn_mask = None
+
+        # partition windows
+        x_windows = window_partition(
+            shifted_x, self.window_size
+        )  # nW*B, window_size, window_size, C
+        x_windows = x_windows.view(
+            -1, self.window_size * self.window_size, C
+        )  # nW*B, window_size*window_size, C
+
+        # W-MSA/SW-MSA
+        attn_windows = self.attn(x_windows, mask=attn_mask)  # nW*B, window_size*window_size, C
+
+        # merge windows
+        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
+        shifted_x = window_reverse(attn_windows, self.window_size, Hp, Wp)  # B H' W' C
+
+        # reverse cyclic shift
+        if self.shift_size > 0:
+            x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
+        else:
+            x = shifted_x
+
+        if pad_r > 0 or pad_b > 0:
+            x = x[:, :H, :W, :].contiguous()
+
+        x = x.view(B, H * W, C)
+
+        # FFN
+        x = shortcut + self.drop_path(x)
+        x = x + self.drop_path(self.mlp(self.norm2(x)))
+
+        return x
+
+
+class PatchMerging(nn.Module):
+    """Patch Merging Layer
+    Args:
+        dim (int): Number of input channels.
+        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
+    """
+
+    def __init__(self, dim, norm_layer=nn.LayerNorm):
+        super().__init__()
+        self.dim = dim
+        self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
+        self.norm = norm_layer(4 * dim)
+
+    def forward(self, x, H, W):
+        """Forward function.
+        Args:
+            x: Input feature, tensor size (B, H*W, C).
+            H, W: Spatial resolution of the input feature.
+        """
+        B, L, C = x.shape
+        assert L == H * W, "input feature has wrong size"
+
+        x = x.view(B, H, W, C)
+
+        # padding
+        pad_input = (H % 2 == 1) or (W % 2 == 1)
+        if pad_input:
+            x = F.pad(x, (0, 0, 0, W % 2, 0, H % 2))
+
+        x0 = x[:, 0::2, 0::2, :]  # B H/2 W/2 C
+        x1 = x[:, 1::2, 0::2, :]  # B H/2 W/2 C
+        x2 = x[:, 0::2, 1::2, :]  # B H/2 W/2 C
+        x3 = x[:, 1::2, 1::2, :]  # B H/2 W/2 C
+        x = torch.cat([x0, x1, x2, x3], -1)  # B H/2 W/2 4*C
+        x = x.view(B, -1, 4 * C)  # B H/2*W/2 4*C
+
+        x = self.norm(x)
+        x = self.reduction(x)
+
+        return x
+
+
+class BasicLayer(nn.Module):
+    """A basic Swin Transformer layer for one stage.
+    Args:
+        dim (int): Number of feature channels
+        depth (int): Depths of this stage.
+        num_heads (int): Number of attention head.
+        window_size (int): Local window size. Default: 7.
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4.
+        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
+        drop (float, optional): Dropout rate. Default: 0.0
+        attn_drop (float, optional): Attention dropout rate. Default: 0.0
+        drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0
+        norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
+        downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None
+        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
+    """
+
+    def __init__(
+        self,
+        dim,
+        depth,
+        num_heads,
+        window_size=7,
+        mlp_ratio=4.0,
+        qkv_bias=True,
+        qk_scale=None,
+        drop=0.0,
+        attn_drop=0.0,
+        drop_path=0.0,
+        norm_layer=nn.LayerNorm,
+        downsample=None,
+        use_checkpoint=False,
+    ):
+        super().__init__()
+        self.window_size = window_size
+        self.shift_size = window_size // 2
+        self.depth = depth
+        self.use_checkpoint = use_checkpoint
+
+        # build blocks
+        self.blocks = nn.ModuleList(
+            [
+                SwinTransformerBlock(
+                    dim=dim,
+                    num_heads=num_heads,
+                    window_size=window_size,
+                    shift_size=0 if (i % 2 == 0) else window_size // 2,
+                    mlp_ratio=mlp_ratio,
+                    qkv_bias=qkv_bias,
+                    qk_scale=qk_scale,
+                    drop=drop,
+                    attn_drop=attn_drop,
+                    drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,
+                    norm_layer=norm_layer,
+                )
+                for i in range(depth)
+            ]
+        )
+
+        # patch merging layer
+        if downsample is not None:
+            self.downsample = downsample(dim=dim, norm_layer=norm_layer)
+        else:
+            self.downsample = None
+
+    def forward(self, x, H, W):
+        """Forward function.
+        Args:
+            x: Input feature, tensor size (B, H*W, C).
+            H, W: Spatial resolution of the input feature.
+        """
+
+        # calculate attention mask for SW-MSA
+        Hp = int(np.ceil(H / self.window_size)) * self.window_size
+        Wp = int(np.ceil(W / self.window_size)) * self.window_size
+        img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device)  # 1 Hp Wp 1
+        h_slices = (
+            slice(0, -self.window_size),
+            slice(-self.window_size, -self.shift_size),
+            slice(-self.shift_size, None),
+        )
+        w_slices = (
+            slice(0, -self.window_size),
+            slice(-self.window_size, -self.shift_size),
+            slice(-self.shift_size, None),
+        )
+        cnt = 0
+        for h in h_slices:
+            for w in w_slices:
+                img_mask[:, h, w, :] = cnt
+                cnt += 1
+
+        mask_windows = window_partition(
+            img_mask, self.window_size
+        )  # nW, window_size, window_size, 1
+        mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
+        attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
+        attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(
+            attn_mask == 0, float(0.0)
+        )
+
+        for blk in self.blocks:
+            blk.H, blk.W = H, W
+            if self.use_checkpoint:
+                x = checkpoint.checkpoint(blk, x, attn_mask)
+            else:
+                x = blk(x, attn_mask)
+        if self.downsample is not None:
+            x_down = self.downsample(x, H, W)
+            Wh, Ww = (H + 1) // 2, (W + 1) // 2
+            return x, H, W, x_down, Wh, Ww
+        else:
+            return x, H, W, x, H, W
+
+
+class PatchEmbed(nn.Module):
+    """Image to Patch Embedding
+    Args:
+        patch_size (int): Patch token size. Default: 4.
+        in_chans (int): Number of input image channels. Default: 3.
+        embed_dim (int): Number of linear projection output channels. Default: 96.
+        norm_layer (nn.Module, optional): Normalization layer. Default: None
+    """
+
+    def __init__(self, patch_size=4, in_chans=3, embed_dim=96, norm_layer=None):
+        super().__init__()
+        patch_size = to_2tuple(patch_size)
+        self.patch_size = patch_size
+
+        self.in_chans = in_chans
+        self.embed_dim = embed_dim
+
+        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size)
+        if norm_layer is not None:
+            self.norm = norm_layer(embed_dim)
+        else:
+            self.norm = None
+
+    def forward(self, x):
+        """Forward function."""
+        # padding
+        _, _, H, W = x.size()
+        if W % self.patch_size[1] != 0:
+            x = F.pad(x, (0, self.patch_size[1] - W % self.patch_size[1]))
+        if H % self.patch_size[0] != 0:
+            x = F.pad(x, (0, 0, 0, self.patch_size[0] - H % self.patch_size[0]))
+
+        x = self.proj(x)  # B C Wh Ww
+        if self.norm is not None:
+            Wh, Ww = x.size(2), x.size(3)
+            x = x.flatten(2).transpose(1, 2)
+            x = self.norm(x)
+            x = x.transpose(1, 2).view(-1, self.embed_dim, Wh, Ww)
+
+        return x
+
+
+class SwinTransformer(nn.Module):
+    """Swin Transformer backbone.
+        A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows`  -
+          https://arxiv.org/pdf/2103.14030
+    Args:
+        pretrain_img_size (int): Input image size for training the pretrained model,
+            used in absolute postion embedding. Default 224.
+        patch_size (int | tuple(int)): Patch size. Default: 4.
+        in_chans (int): Number of input image channels. Default: 3.
+        embed_dim (int): Number of linear projection output channels. Default: 96.
+        depths (tuple[int]): Depths of each Swin Transformer stage.
+        num_heads (tuple[int]): Number of attention head of each stage.
+        window_size (int): Window size. Default: 7.
+        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4.
+        qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True
+        qk_scale (float): Override default qk scale of head_dim ** -0.5 if set.
+        drop_rate (float): Dropout rate.
+        attn_drop_rate (float): Attention dropout rate. Default: 0.
+        drop_path_rate (float): Stochastic depth rate. Default: 0.2.
+        norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm.
+        ape (bool): If True, add absolute position embedding to the patch embedding. Default: False.
+        patch_norm (bool): If True, add normalization after patch embedding. Default: True.
+        out_indices (Sequence[int]): Output from which stages.
+        frozen_stages (int): Stages to be frozen (stop grad and set eval mode).
+            -1 means not freezing any parameters.
+        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
+        dilation (bool): if True, the output size if 16x downsample, ow 32x downsample.
+    """
+
+    def __init__(
+        self,
+        pretrain_img_size=224,
+        patch_size=4,
+        in_chans=3,
+        embed_dim=96,
+        depths=[2, 2, 6, 2],
+        num_heads=[3, 6, 12, 24],
+        window_size=7,
+        mlp_ratio=4.0,
+        qkv_bias=True,
+        qk_scale=None,
+        drop_rate=0.0,
+        attn_drop_rate=0.0,
+        drop_path_rate=0.2,
+        norm_layer=nn.LayerNorm,
+        ape=False,
+        patch_norm=True,
+        out_indices=(0, 1, 2, 3),
+        frozen_stages=-1,
+        dilation=False,
+        use_checkpoint=False,
+    ):
+        super().__init__()
+
+        self.pretrain_img_size = pretrain_img_size
+        self.num_layers = len(depths)
+        self.embed_dim = embed_dim
+        self.ape = ape
+        self.patch_norm = patch_norm
+        self.out_indices = out_indices
+        self.frozen_stages = frozen_stages
+        self.dilation = dilation
+
+        # if use_checkpoint:
+        #     print("use_checkpoint!!!!!!!!!!!!!!!!!!!!!!!!")
+
+        # split image into non-overlapping patches
+        self.patch_embed = PatchEmbed(
+            patch_size=patch_size,
+            in_chans=in_chans,
+            embed_dim=embed_dim,
+            norm_layer=norm_layer if self.patch_norm else None,
+        )
+
+        # absolute position embedding
+        if self.ape:
+            pretrain_img_size = to_2tuple(pretrain_img_size)
+            patch_size = to_2tuple(patch_size)
+            patches_resolution = [
+                pretrain_img_size[0] // patch_size[0],
+                pretrain_img_size[1] // patch_size[1],
+            ]
+
+            self.absolute_pos_embed = nn.Parameter(
+                torch.zeros(1, embed_dim, patches_resolution[0], patches_resolution[1])
+            )
+            trunc_normal_(self.absolute_pos_embed, std=0.02)
+
+        self.pos_drop = nn.Dropout(p=drop_rate)
+
+        # stochastic depth
+        dpr = [
+            x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))
+        ]  # stochastic depth decay rule
+
+        # build layers
+        self.layers = nn.ModuleList()
+        # prepare downsample list
+        downsamplelist = [PatchMerging for i in range(self.num_layers)]
+        downsamplelist[-1] = None
+        num_features = [int(embed_dim * 2**i) for i in range(self.num_layers)]
+        if self.dilation:
+            downsamplelist[-2] = None
+            num_features[-1] = int(embed_dim * 2 ** (self.num_layers - 1)) // 2
+        for i_layer in range(self.num_layers):
+            layer = BasicLayer(
+                # dim=int(embed_dim * 2 ** i_layer),
+                dim=num_features[i_layer],
+                depth=depths[i_layer],
+                num_heads=num_heads[i_layer],
+                window_size=window_size,
+                mlp_ratio=mlp_ratio,
+                qkv_bias=qkv_bias,
+                qk_scale=qk_scale,
+                drop=drop_rate,
+                attn_drop=attn_drop_rate,
+                drop_path=dpr[sum(depths[:i_layer]) : sum(depths[: i_layer + 1])],
+                norm_layer=norm_layer,
+                # downsample=PatchMerging if (i_layer < self.num_layers - 1) else None,
+                downsample=downsamplelist[i_layer],
+                use_checkpoint=use_checkpoint,
+            )
+            self.layers.append(layer)
+
+        # num_features = [int(embed_dim * 2 ** i) for i in range(self.num_layers)]
+        self.num_features = num_features
+
+        # add a norm layer for each output
+        for i_layer in out_indices:
+            layer = norm_layer(num_features[i_layer])
+            layer_name = f"norm{i_layer}"
+            self.add_module(layer_name, layer)
+
+        self._freeze_stages()
+
+    def _freeze_stages(self):
+        if self.frozen_stages >= 0:
+            self.patch_embed.eval()
+            for param in self.patch_embed.parameters():
+                param.requires_grad = False
+
+        if self.frozen_stages >= 1 and self.ape:
+            self.absolute_pos_embed.requires_grad = False
+
+        if self.frozen_stages >= 2:
+            self.pos_drop.eval()
+            for i in range(0, self.frozen_stages - 1):
+                m = self.layers[i]
+                m.eval()
+                for param in m.parameters():
+                    param.requires_grad = False
+
+    # def init_weights(self, pretrained=None):
+    #     """Initialize the weights in backbone.
+    #     Args:
+    #         pretrained (str, optional): Path to pre-trained weights.
+    #             Defaults to None.
+    #     """
+
+    #     def _init_weights(m):
+    #         if isinstance(m, nn.Linear):
+    #             trunc_normal_(m.weight, std=.02)
+    #             if isinstance(m, nn.Linear) and m.bias is not None:
+    #                 nn.init.constant_(m.bias, 0)
+    #         elif isinstance(m, nn.LayerNorm):
+    #             nn.init.constant_(m.bias, 0)
+    #             nn.init.constant_(m.weight, 1.0)
+
+    #     if isinstance(pretrained, str):
+    #         self.apply(_init_weights)
+    #         logger = get_root_logger()
+    #         load_checkpoint(self, pretrained, strict=False, logger=logger)
+    #     elif pretrained is None:
+    #         self.apply(_init_weights)
+    #     else:
+    #         raise TypeError('pretrained must be a str or None')
+
+    def forward_raw(self, x):
+        """Forward function."""
+        x = self.patch_embed(x)
+
+        Wh, Ww = x.size(2), x.size(3)
+        if self.ape:
+            # interpolate the position embedding to the corresponding size
+            absolute_pos_embed = F.interpolate(
+                self.absolute_pos_embed, size=(Wh, Ww), mode="bicubic"
+            )
+            x = (x + absolute_pos_embed).flatten(2).transpose(1, 2)  # B Wh*Ww C
+        else:
+            x = x.flatten(2).transpose(1, 2)
+        x = self.pos_drop(x)
+
+        outs = []
+        for i in range(self.num_layers):
+            layer = self.layers[i]
+            x_out, H, W, x, Wh, Ww = layer(x, Wh, Ww)
+            # import ipdb; ipdb.set_trace()
+
+            if i in self.out_indices:
+                norm_layer = getattr(self, f"norm{i}")
+                x_out = norm_layer(x_out)
+
+                out = x_out.view(-1, H, W, self.num_features[i]).permute(0, 3, 1, 2).contiguous()
+                outs.append(out)
+        # in:
+        #   torch.Size([2, 3, 1024, 1024])
+        # outs:
+        #   [torch.Size([2, 192, 256, 256]), torch.Size([2, 384, 128, 128]), \
+        #       torch.Size([2, 768, 64, 64]), torch.Size([2, 1536, 32, 32])]
+        return tuple(outs)
+
+    def forward(self, tensor_list: NestedTensor):
+        x = tensor_list.tensors
+
+        """Forward function."""
+        x = self.patch_embed(x)
+
+        Wh, Ww = x.size(2), x.size(3)
+        if self.ape:
+            # interpolate the position embedding to the corresponding size
+            absolute_pos_embed = F.interpolate(
+                self.absolute_pos_embed, size=(Wh, Ww), mode="bicubic"
+            )
+            x = (x + absolute_pos_embed).flatten(2).transpose(1, 2)  # B Wh*Ww C
+        else:
+            x = x.flatten(2).transpose(1, 2)
+        x = self.pos_drop(x)
+
+        outs = []
+        for i in range(self.num_layers):
+            layer = self.layers[i]
+            x_out, H, W, x, Wh, Ww = layer(x, Wh, Ww)
+
+            if i in self.out_indices:
+                norm_layer = getattr(self, f"norm{i}")
+                x_out = norm_layer(x_out)
+
+                out = x_out.view(-1, H, W, self.num_features[i]).permute(0, 3, 1, 2).contiguous()
+                outs.append(out)
+        # in:
+        #   torch.Size([2, 3, 1024, 1024])
+        # out:
+        #   [torch.Size([2, 192, 256, 256]), torch.Size([2, 384, 128, 128]), \
+        #       torch.Size([2, 768, 64, 64]), torch.Size([2, 1536, 32, 32])]
+
+        # collect for nesttensors
+        outs_dict = {}
+        for idx, out_i in enumerate(outs):
+            m = tensor_list.mask
+            assert m is not None
+            mask = F.interpolate(m[None].float(), size=out_i.shape[-2:]).to(torch.bool)[0]
+            outs_dict[idx] = NestedTensor(out_i, mask)
+
+        return outs_dict
+
+    def train(self, mode=True):
+        """Convert the model into training mode while keep layers freezed."""
+        super(SwinTransformer, self).train(mode)
+        self._freeze_stages()
+
+
+def build_swin_transformer(modelname, pretrain_img_size, **kw):
+    assert modelname in [
+        "swin_T_224_1k",
+        "swin_B_224_22k",
+        "swin_B_384_22k",
+        "swin_L_224_22k",
+        "swin_L_384_22k",
+    ]
+
+    model_para_dict = {
+        "swin_T_224_1k": dict(
+            embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24], window_size=7
+        ),
+        "swin_B_224_22k": dict(
+            embed_dim=128, depths=[2, 2, 18, 2], num_heads=[4, 8, 16, 32], window_size=7
+        ),
+        "swin_B_384_22k": dict(
+            embed_dim=128, depths=[2, 2, 18, 2], num_heads=[4, 8, 16, 32], window_size=12
+        ),
+        "swin_L_224_22k": dict(
+            embed_dim=192, depths=[2, 2, 18, 2], num_heads=[6, 12, 24, 48], window_size=7
+        ),
+        "swin_L_384_22k": dict(
+            embed_dim=192, depths=[2, 2, 18, 2], num_heads=[6, 12, 24, 48], window_size=12
+        ),
+    }
+    kw_cgf = model_para_dict[modelname]
+    kw_cgf.update(kw)
+    model = SwinTransformer(pretrain_img_size=pretrain_img_size, **kw_cgf)
+    return model
+
+
+if __name__ == "__main__":
+    model = build_swin_transformer("swin_L_384_22k", 384, dilation=True)
+    x = torch.rand(2, 3, 1024, 1024)
+    y = model.forward_raw(x)
+    import ipdb
+
+    ipdb.set_trace()
+    x = torch.rand(2, 3, 384, 384)
+    y = model.forward_raw(x)