Learning with Unreliability: Fast Few-shot Voxel Radiance Fields with Relative Geometric Consistency (CVPR2024)

paper

git clone https://github.com/HKCLynn/ReVoRF.git
cd ReVoRF
pip install -r requirements.txt

Pytorch and torch_scatter installation is machine dependent.

NeRF Synthetic Dataset and LLFF Dataset

data
├── nerf_synthetic     
│   └── [chair|drums|ficus|hotdog|lego|materials|mic|ship]
│       ├── [train|val|test]
│       │   └── r_*.png
│       └── transforms_[train|val|test].json
│
├── nerf_llff_data     
    └── [fern|flower|fortress|horns|leaves|orchids|room|trex]

• Get Depth Maps

  If the datasets are set in the above formats. Get DPT model and generate the depth maps.

   $ python download.py

• Training

  $ python run.py --config configs/nerf/hotdog.py --render_test

• Evaluation

  $ python run.py --config configs/nerf/hotdog.py --render_only --render_test \
                                                --eval_ssim --eval_lpips_vgg

The code is heavily based on DVGOv2 implementation, and some functions are modified from VGOS. Thank you!