Creating Custom Dataset

By default Nerf.jl and NerfGUI.jl comes with its own Raccoon dataset.

However, it is easy to create custom ones.

Steps

Nerf.jl uses same input format as instant-ngp.

1. Take a short clip of an object you want to reconstruct.

2. Using colmap2nerf.py script reconstruct camera poses via COLMAP. Below is an example command:

   ~$ python3 colmap2nerf.py --video_in=input.mp4 --video_fps=2 --run_colmap --aabb_scale=4

   It will slice video into frames, run COLMAP camera pose reconstruction and save the results in transforms.json file along with images in images/ directory.

   Make sure paths in transform.json point correctly to the images.

3. Open transforms.json in NerfGUI.jl application by pasting its path in the datasets window.