Cameras as Rays

[arXiv] [Project Page] [Bibtex] [Colab]

This repository contains code for "Cameras as Rays: Pose Estimation via Ray Diffusion" (ICLR 2024).

Clone the repository:

git clone --depth=1 --branch=main https://github.com/jasonyzhang/RayDiffusion.git

Setting up Environment

We recommend using a conda environment to manage dependencies. Install a version of Pytorch compatible with your CUDA version from the Pytorch website.

conda create -n raydiffusion python=3.10
conda activate raydiffusion
conda install pytorch==2.1.1 torchvision==0.16.1 torchaudio==2.1.1 pytorch-cuda=11.8 -c pytorch -c nvidia
conda install xformers -c xformers
pip install -r requirements.txt

Then, follow the directions to install Pytorch3D here. We recommend installing Pytorch3D using the pre-built wheel with the corresponding Python/Pytorch/CUDA version:

pip install --no-index --no-cache-dir pytorch3d -f https://dl.fbaipublicfiles.com/pytorch3d/packaging/wheels/py310_cu118_pyt211/download.html

If you are having trouble installing using the pre-built wheel, you can also try building from source, but this will take a lot longer.

Run Demo

Download the model weights from Google Drive.

gdown https://drive.google.com/uc\?id\=1anIKsm66zmDiFuo8Nmm1HupcitM6NY7e
unzip models.zip

Run ray diffusion with known bounding boxes (provided as a json):

python demo.py  --model_dir models/co3d_diffusion --image_dir examples/robot/images \
    --bbox_path examples/robot/bboxes.json --output_path robot.html

Run ray diffusion with bounding boxes extracted automatically from masks:

python demo.py  --model_dir models/co3d_diffusion --image_dir examples/robot/images \
    --mask_dir examples/robot/masks --output_path robot.html

Run ray regression:

python demo.py  --model_dir models/co3d_regression --image_dir examples/robot/images \
    --bbox_path examples/robot/bboxes.json --output_path robot.html

Training

Training command for ray diffusion:

accelerate launch --multi_gpu --gpu_ids 0,1,2,3,4,5,6,7 --num_processes 8 train.py \
    training.batch_size=8 training.max_iterations=450000

See docs/train.md for more detailed instructions on training.

Evaluation

See docs/eval.md for instructions on how to run evaluation code.

Citing Cameras as Rays

If you find this code helpful, please cite:

@InProceedings{zhang2024raydiffusion,
    title={Cameras as Rays: Pose Estimation via Ray Diffusion},
    author={Zhang, Jason Y and Lin, Amy and Kumar, Moneish and Yang, Tzu-Hsuan and Ramanan, Deva and Tulsiani, Shubham},
    booktitle={International Conference on Learning Representations (ICLR)},
    year={2024}
}