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