|
|
--- |
|
|
license: mit |
|
|
--- |
|
|
# Fastmap evaluation suite. |
|
|
|
|
|
|
|
|
You only need the databases to run fastmap. Download the images if you want to produce colored point cloud. |
|
|
Download the subset of data you want to your local directory. |
|
|
```bash |
|
|
huggingface-cli download whc/fastmap_sfm --repo-type dataset --local-dir ./ --include 'databases/tnt_*' 'ground_truths/tnt_*' |
|
|
``` |
|
|
or use the python interface |
|
|
```python |
|
|
from huggingface_hub import hf_hub_download, snapshot_download |
|
|
snapshot_download( |
|
|
repo_id="whc/fastmap_sfm", repo_type='dataset', |
|
|
local_dir="./", |
|
|
allow_patterns=["ground_truths/*",], |
|
|
max_workers=8 |
|
|
) |
|
|
``` |
|
|
|
|
|
## Images |
|
|
* MipNeRF360, Tanks and Temples, NeRF-OSR, DroneDeploy, ZipNeRF and Mill-19 store the images for a scene in the same directory. All the images are used for running SfM. |
|
|
* For Eyeful Tower, images are stored in different subdirectories for each scene. We extract the image paths from the provided `cameras.json`, which contains the GT poses for each image. The subdirectory structure is preserved. |
|
|
* Urbanscene3D also stores images in different subdirectories. We extract the image paths from the refined GT provided [here](https://github.com/cmusatyalab/mega-nerf?tab=readme-ov-file#urbanscene-3d) by [MegaNeRF](https://github.com/cmusatyalab/mega-nerf). |
|
|
|
|
|
## Databases |
|
|
Unless specified below, all databases `.db` files are produced with |
|
|
```bash |
|
|
colmap feature_extractor --ImageReader.single_camera 1 --ImageReader.camera_model SIMPLE_RADIAL --image_path {imgdir} --database_path {db_fname} |
|
|
colmap exhaustive_matcher --database_path {db_fname} |
|
|
``` |
|
|
|
|
|
* For nerf-osr (i.e. those with prefix `nosr_`), `dploy_house4`, urbanscene (`urbn_`), eyeful tower (`eft_`), we use `--ImageReader.single_camera 0`. |
|
|
* For eth3d MVS (`eth3d_dslr_`), we populate the databases with author provided ground truth camera intrinsics. This is to keep the eval protocol consistent with Glomap. see [issue 96](https://github.com/colmap/glomap/issues/96). |
|
|
|
|
|
|
|
|
## Ground truths |
|
|
|
|
|
We use two types of ground-truth formats. The first is colmap `.bin` output. |
|
|
Secondly, for datasets whose ground truths come in non-standard format, we verify and convert them to a white-box json file each containing a list of `{fname: str, c2w: list[float]}`. |
|
|
The `c2w` matrix is in OpenCV convention. |
|
|
|
|
|
We have added our gt convertion script `convert_gt.py` to this repo as a reference. |
|
|
|
|
|
## glomap/colmap container |
|
|
We provide a singularity container `glomap_250121.sif` in the repo, and a docker container [here](https://hub.docker.com/r/haochenw/glomap/tags). |
|
|
The Dockerfile is printed in the dockerhub overview. |
|
|
|
|
|
|
|
|
|
