MultiNeRF

An pytorch reimplementation of multinerf.

News

• I have implemented "Zip-NeRF: Anti-Aliased Grid-Based Neural Radiance Fields" https://arxiv.org/abs/2304.06706, which is available at here

Install

# Clone the repo.
git clone https://https://github.com/SuLvXiangXin/zipnerf-pytorch.git
cd multi-pytorch

# Make a conda environment.
conda create --name multinerf python=3.9
conda activate multinerf

# Install requirements.
pip install -r requirements.txt

Dataset

You can download several datasets below:

MipNeRF360: mipnerf360_data

RefNeRF: refnerf_data

RawNeRF: raw_data

Origin NeRF: nerf_llff_data nerf_synthetic

mkdir data
cd data

# for example mipnerf360_data
wget http://storage.googleapis.com/gresearch/refraw360/360_v2.zip
unzip 360_v2.zip

Train

# Configure your training (DDP? fp16? ...)
# see https://huggingface.co/docs/accelerate/index for details
accelerate config

# Where your data is 
DATA_DIR=data/360_v2/bicycle
EXP_NAME=360_v2/bicycle

# Experiment will be conducted under "exp/${EXP_NAME}" folder
# "--gin_configs=configs/360.gin" can be seen as a default config 
# and you can add specific config useing --gin_bindings="..." 
accelerate launch train.py \
    --gin_configs=configs/360.gin \
    --gin_bindings="Config.data_dir = '${DATA_DIR}'" \
    --gin_bindings="Config.exp_name = '${EXP_NAME}'" \
    --gin_bindings="Config.factor = 4"

# or you can also run without accelerate (without DDP)
CUDA_VISIBLE_DEVICES=0 python train.py \
    --gin_configs=configs/360.gin \
    --gin_bindings="Config.data_dir = '${DATA_DIR}'" \
    --gin_bindings="Config.exp_name = '${EXP_NAME}'" \
      --gin_bindings="Config.factor = 4" 

# alternative you can use an example training script 
bash script/train_360.sh

# metric, render image, etc can be viewed through tensorboard
tensorboard --logdir "exp/${EXP_NAME}"

Render

Rendering results can be found in the directory exp/${EXP_NAME}/render

accelerate launch render.py \
    --gin_configs=configs/360.gin \
    --gin_bindings="Config.data_dir = '${DATA_DIR}'" \
    --gin_bindings="Config.exp_name = '${EXP_NAME}'" \
    --gin_bindings="Config.render_path = True" \
    --gin_bindings="Config.render_path_frames = 480" \
    --gin_bindings="Config.render_video_fps = 60" \
    --gin_bindings="Config.factor = 4"  

# alternative you can use an example rendering script 
bash script/render_360.sh

Evaluate

Evaluating results can be found in the directory exp/${EXP_NAME}/test_preds

# using the same exp_name as in training
accelerate launch eval.py \
    --gin_configs=configs/360.gin \
    --gin_bindings="Config.data_dir = '${DATA_DIR}'" \
    --gin_bindings="Config.exp_name = '${EXP_NAME}'" \
    --gin_bindings="Config.factor = 4"


# alternative you can use an example evaluating script 
bash script/eval_360.sh

OutOfMemory

you can decrease the total batch size by adding e.g. --gin_bindings="Config.batch_size = 4096" , or decrease the test chunk size by adding e.g. --gin_bindings="Config.render_chunk_size = 4096" , or use more GPU by configure accelerate config .

Preparing custom data

More details can be found at https://github.com/google-research/multinerf

DATA_DIR=my_dataset_dir
bash scripts/local_colmap_and_resize.sh ${DATA_DIR}

Citation

@misc{multinerf2022,
      title={{MultiNeRF}: {A} {Code} {Release} for {Mip-NeRF} 360, {Ref-NeRF}, and {RawNeRF}},
      author={Ben Mildenhall and Dor Verbin and Pratul P. Srinivasan and Peter Hedman and Ricardo Martin-Brualla and Jonathan T. Barron},
      year={2022},
      url={https://github.com/google-research/multinerf},
}

@Misc{accelerate,
  title =        {Accelerate: Training and inference at scale made simple, efficient and adaptable.},
  author =       {Sylvain Gugger, Lysandre Debut, Thomas Wolf, Philipp Schmid, Zachary Mueller, Sourab Mangrulkar},
  howpublished = {\url{https://github.com/huggingface/accelerate}},
  year =         {2022}
}

Acknowledgements

• Thanks to multinerf for amazing multinerf(MipNeRF360,RefNeRF,RawNeRF) implementation
• Thanks to accelerate for distributed training