data_guide.md

Data Convention

From Images

If you have a set of multiple view images, you need to run COLMAP to estimate the camera parameters. You can do it in COLMAP GUI or try data/img2poses.py. You may also refer to nerf_pl for more explanation on this step.

Additionally, if you want to reconstruction with NeuS, you might need to extract the foreground matte from the images. This could be achieve by various methods. We provide a colab notebook [here].

From Colmap

If you have the camera parameters estimated by COLMAP with structure like following, you need to convert the COLMAP format to VolSDF format and NeuS format respectively.

scene/
├── sparse_points.ply
├── images/
├── sparse/
    ├── 0/
        ├── cameras.bin
        ├── images.bin
        ├── points3D.bin

to VolSDF format

Try following command:

python colmap2volsdf.py --root_dir $YOUR_SCENE_DIR 

For a more detailed usage, please refer to VolSDF's data convention.

to NeuS format

Please refer to NeuS's data convention.