Neural Invertible Warp for NeRF (ECCV-2024)

June 29, 2025 · View on GitHub

This is an official implementation of the paper "Neural Invertible Warp for NeRF".

🛠️ Installation Steps

Assuming a fresh Anaconda environment, you can install the dependencies by

pip install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 --index-url https://download.pytorch.org/whl/cu118
numpy==1.26.4

pip install -r requirements.txt

conda install -c conda-forge cupy cuda-version=11.6

💿 Training data

DTU

Images: We use the DTU dataset, produced by SPARF, where the images are processed and resized to 300 x 400. Download the data here.
Ground-truth depth maps: For geometry evaluation, we report the depth error. Download the depth maps. They are extracted from MVSNeRF.

LLFF

The LLFF real-world data can be found in the NeRF Google Drive. You can download the dataset by running

gdown 16VnMcF1KJYxN9QId6TClMsZRahHNMW5g # download nerf_llff_data.zip
unzip nerf_llff_data.zip
rm -f nerf_llff_data.zip
mv nerf_llff_data data/llff

The data directory should contain the subdirectories llff and dtu. If you have downloaded the datasets, you can create soft links to them within the data directory.

⏳ Training and Evaluation

All checkpoints, logs, and Tensorboard event files are written to ~/output/<GROUP>/<NAME> by default. Override this with --output_root=/your_path.

Quick start

#   Train the reference model
bash scripts/train_llff.sh

#   Evaluate the released checkpoint
bash scripts/eval_llff.sh

Dataset	Training script	Eval script	Notes
LLFF	`train_llff.sh`	`eval_llff.sh`	Matches Table 1 in supplementary paper
DTU	`train_dtu.sh`	`eval_dtu.sh`	Matches Table 2 in supplementary paper

👩‍💻 Citation

This code is for non-commercial use. If you find our work useful in your research please cite our paper:

@inproceedings{chng2024invertible,
  title={Invertible neural warp for nerf},
  author={Chng, Shin-Fang and Garg, Ravi and Saratchandran, Hemanth and Lucey, Simon},
  booktitle={European Conference on Computer Vision},
  pages={405--421},
  year={2024},
  organization={Springer}
}