Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
@@ -1,3 +1,126 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
<div align="center">
|
2 |
+
<h2 align="center">Pixels, Points, Polygons: A Global Dataset and Baseline for Multimodal Building Vectorization</h2>
|
3 |
+
<!-- <h3 align="center">Arxiv</h3> -->
|
4 |
+
<!-- <h3 align="center"><a href="https://raphaelsulzer.de/">Raphael Sulzer<sup>1,2</sup></a><br></h3> -->
|
5 |
+
<h3><align="center">Raphael Sulzer<sup>1,2</sup></a></h3>
|
6 |
+
<align="center"><sup>1</sup>LuxCarta <sup>2</sup>Inria
|
7 |
+
<img src="./media/teaser.jpg" width=100% height=100%>
|
8 |
+
<b>Figure 1</b>: A view of our dataset of Zurich, Switzerland
|
9 |
+
</div>
|
10 |
+
|
11 |
+
|
12 |
+
<!-- [[Project Webpage]()] [[Paper](https://arxiv.org/abs/2412.07899)] [[Video]()] -->
|
13 |
+
|
14 |
+
## Abstract:
|
15 |
+
|
16 |
+
asd
|
17 |
+
|
18 |
+
## Highlights
|
19 |
+
|
20 |
+
- A global, multimodal dataset of aerial images, aerial lidar point clouds and building polygons
|
21 |
+
- A library for training and evaluating state-of-the-art deep learning methods on the dataset
|
22 |
+
|
23 |
+
|
24 |
+
## Dataset
|
25 |
+
|
26 |
+
### Numbers
|
27 |
+
|
28 |
+
#TODO put some images and numbers about the dataset
|
29 |
+
|
30 |
+
<!-- ### Properties -->
|
31 |
+
|
32 |
+
We provide train and val splits of the dataset in two different sizes 224 $\times$ 224 and 512 $\times$ 512. Both sized versions cover the same areas. The tiles of the test split have a fixed size of 2000 $\times$ 2000.
|
33 |
+
|
34 |
+
### Download
|
35 |
+
|
36 |
+
hugginface link
|
37 |
+
|
38 |
+
### Prepare custom tile size
|
39 |
+
|
40 |
+
See [datasets preprocessing](data_preprocess) for instructions on preparing a dataset with different tile sizes.
|
41 |
+
|
42 |
+
|
43 |
+
## Requirements
|
44 |
+
|
45 |
+
To create a conda environment named `ppp` and install the repository as a python package with all dependencies run
|
46 |
+
```
|
47 |
+
bash install.sh
|
48 |
+
```
|
49 |
+
|
50 |
+
or, if you want to manage the environment yourself run
|
51 |
+
```
|
52 |
+
pip install -r requirements-torch-cuda.txt
|
53 |
+
pip install .
|
54 |
+
```
|
55 |
+
β οΈ **Warning**: The implementation of the LiDAR point cloud encoder uses Open3D-ML. Currently, Open3D-ML officially only supports the PyTorch version specified in `requirements-torch-cuda.txt`.
|
56 |
+
|
57 |
+
|
58 |
+
|
59 |
+
## Model Zoo
|
60 |
+
|
61 |
+
|
62 |
+
| Model | \<model> | Encoder | \<encoder> |Image |LiDAR | IoU | C-IoU |
|
63 |
+
|--------------- |---- |--------------- |--------------- |--- |--- |----- |----- |
|
64 |
+
| Frame Field Learning |\<ffl> | Vision Transformer (ViT) | \<vit_cnn> | β
| | 0.85 | 0.90 |
|
65 |
+
| Frame Field Learning |\<ffl> | PointPillars (PP) + ViT | \<pp_vit_cnn> | | β
| 0.80 | 0.88 |
|
66 |
+
| Frame Field Learning |\<ffl> | PP+ViT \& ViT | \<fusion_vit_cnn> | β
|β
| 0.78 | 0.85 |
|
67 |
+
| HiSup |\<hisup> | Vision Transformer (ViT) | \<vit_cnn> | β
| | 0.85 | 0.90 |
|
68 |
+
| HiSup |\<hisup> | PointPillars (PP) + ViT | \<pp_vit_cnn> | | β
| 0.80 | 0.88 |
|
69 |
+
| HiSup |\<hisup> | PP+ViT \& ViT | \<fusion_vit> | β
|β
| 0.78 | 0.85 |
|
70 |
+
| Pix2Poly |\<pix2poly>| Vision Transformer (ViT) | \<vit> | β
| | 0.85 | 0.90 |
|
71 |
+
| Pix2Poly |\<pix2poly>| PointPillars (PP) + ViT | \<pp_vit> | | β
| 0.80 | 0.88 |
|
72 |
+
| Pix2Poly |\<pix2poly>| PP+ViT \& ViT | \<fusion_vit> | β
|β
| 0.78 | 0.85 |
|
73 |
+
|
74 |
+
## Configuration
|
75 |
+
|
76 |
+
The project supports hydra configuration which allows to modify any parameter from the command line, such as the model and encoder types from the table above.
|
77 |
+
To view all available options run
|
78 |
+
```
|
79 |
+
python train.py --help
|
80 |
+
```
|
81 |
+
|
82 |
+
## Training
|
83 |
+
|
84 |
+
Start training with the following command:
|
85 |
+
|
86 |
+
```
|
87 |
+
torchrun --nproc_per_node=<num GPUs> train.py model=<model> encoder=<encoder> model.batch_size=<batch size> ...
|
88 |
+
|
89 |
+
```
|
90 |
+
|
91 |
+
## Prediction
|
92 |
+
|
93 |
+
```
|
94 |
+
torchrun --nproc_per_node=<num GPUs> predict.py model=<model> checkpoint=best_val_iou ...
|
95 |
+
|
96 |
+
```
|
97 |
+
|
98 |
+
## Evaluation
|
99 |
+
|
100 |
+
```
|
101 |
+
python evaluate.py model=<model> checkpoint=best_val_iou
|
102 |
+
```
|
103 |
+
<!-- ## Trained models
|
104 |
+
|
105 |
+
asd -->
|
106 |
+
|
107 |
+
|
108 |
+
<!-- ## Results
|
109 |
+
|
110 |
+
#TODO Put paper main results table here -->
|
111 |
+
|
112 |
+
|
113 |
+
## Citation
|
114 |
+
|
115 |
+
If you find our work useful, please consider citing:
|
116 |
+
```bibtex
|
117 |
+
...
|
118 |
+
```
|
119 |
+
|
120 |
+
## Acknowledgements
|
121 |
+
|
122 |
+
This repository benefits from the following open-source work. We thank the authors for their great work.
|
123 |
+
|
124 |
+
1. [Frame Field Learning](https://github.com/Lydorn/Polygonization-by-Frame-Field-Learning)
|
125 |
+
2. [HiSup](https://github.com/SarahwXU/HiSup)
|
126 |
+
3. [Pix2Poly](https://github.com/yeshwanth95/Pix2Poly)
|