Skip to content

Latest commit

 

History

History
132 lines (104 loc) · 4.53 KB

README.md

File metadata and controls

132 lines (104 loc) · 4.53 KB

F2-NeRF

This is the repo for the implementation of F2-NeRF: Fast Neural Radiance Field Training with Free Camera Trajectories.

Install

The development of this project is primarily based on LibTorch.

Step 1. Install dependencies

For Debian based Linux distributions:

sudo apt install zlib1g-dev

For Arch based Linux distributions:

sudo pacman -S zlib

Step 2. Clone this repository

git clone --recursive https://github.com/Totoro97/f2-nerf.git
cd f2-nerf

Step 3. Download pre-compiled LibTorch

We take torch-1.13.1+cu117 for example.

cd External
wget https://download.pytorch.org/libtorch/cu117/libtorch-cxx11-abi-shared-with-deps-1.13.1%2Bcu117.zip
unzip ./libtorch-cxx11-abi-shared-with-deps-1.13.1%2Bcu117.zip

Step 4. Compile

The lowest g++ version I have tested is 7.5.0.

cd ..
cmake . -B build
cmake --build build --target main --config RelWithDebInfo -j

Run

Training

Here is an example command to train F2-NeRF:

python scripts/run.py --config-name=wanjinyou dataset_name=example case_name=ngp_fox mode=train +work_dir=$(pwd)

Render test images

Simply run:

python scripts/run.py --config-name=wanjinyou dataset_name=example case_name=ngp_fox mode=test is_continue=true +work_dir=$(pwd)

Render path

We provide a script to generate render path (by interpolating the input camera poses). For example, for the fox data, run:

python scripts/inter_poses.py --data_dir ./data/example/ngp_fox --key_poses 5,10,15,20,25,30,35,40,45,49 --n_out_poses 200

The file poses_render.npy in the data directory would be generated. Then run

python scripts/run.py --config-name=wanjinyou dataset_name=example case_name=ngp_fox mode=render_path is_continue=true +work_dir=$(pwd)

The synthesized images can be found in ./exp/ngp_fox/test/novel_images.

Train F2-NeRF on your custom data

Make sure the images are at ./data/<your-dataset-name>/<your-case-name>/images

  1. Run COLMAP SfM:
bash scripts/local_colmap_and_resize.sh ./data/<your-dataset-name>/<your-case-name>

or run hloc if COLMAP failed. (Make sure hloc has been installed)

bash scripts/local_hloc_and_resize.sh ./data/<your-dataset-name>/<your-case-name>
  1. Generate cameras file:
python scripts/colmap2poses.py --data_dir ./data/<your-dataset-name>/<your-case-name>
  1. Run F2-NeRF using the similar command as in the example data:
python scripts/run.py --config-name=wanjinyou \
dataset_name=<your-dataset-name> case_name=<your-case-name> mode=train \
+work_dir=$(pwd)

Train F2-NeRF on LLFF/NeRF-360-V2 dataset

We provide a script to convert the LLFF camera format to our camera format. For example:

python scripts/llff2poses.py --data_dir=xxx/nerf_llff_data/horns

TODO/Future work

  • Add anti-aliasing

Acknowledgment

Besides LibTorch, this project is also built upon the following awesome libraries:

Some of the code snippets are inspired from instant-ngp, torch-ngp and ngp-pl. The COLMAP processing scripts are from multinerf. The example data ngp_fox is from instant-ngp.

Citation

Cite as below if you find this repository is helpful to your project:

@article{wang2023f2nerf,
  title={F2-NeRF: Fast Neural Radiance Field Training with Free Camera Trajectories},
  author={Wang, Peng and Liu, Yuan and Chen, Zhaoxi and Liu, Lingjie and Liu, Ziwei and Komura, Taku and Theobalt, Christian and Wang, Wenping},
  journal={CVPR},
  year={2023}
}