Official implementation of the ECCV 2022 paper: Contact, Human and Object REconstruction from a single RGB image
News: our new CVPR'23 paper that can track human and object consistently from monocular videos. Check out here!
The code is tested with torch 1.6, cuda10.1, debian 11. We recommend using anaconda environment:
conda create -n chore python=3.7
conda activate chore Main dependencies are listed in requirements.txt, install them with
git clone https://github.com/xiexh20/CHORE.git && cd CHORE
pip install -r requirements.txtInstalling other dependencies:
- psbody-mesh library. see installation.
- igl library.
conda install -c conda-forge igl - Detectron2 library:
python -m pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu101/torch1.6/index.html- Neural Mesh Renderer:
pip install external/neural_renderer - Mesh intersection library:
export CHORE_PATH=${PWD}
git clone https://github.com/NVIDIA/cuda-samples.git external/cuda-samples
export CUDA_SAMPLES_INC=${CHORE_PATH}/external/cuda-samples/Common/
git clone https://github.com/vchoutas/torch-mesh-isect external/torch-mesh-isect
cp external/torch-mesh-isect/include/double_vec_ops.h external/torch-mesh-isect/src/Add these lines to external/torch-mesh-isect/src/bvh.cpp before AT_CHECK is defined (reference):
#ifndef AT_CHECK
#define AT_CHECK TORCH_CHECK
#endif finally run pip install external/torch-mesh-isect/
Pretrained model can be downloaded from here. Please download and extract it in the code directory with: unzip chore-pretrained.zip -d experiments
We use the SMPL-H body model, please prepare it from the official website and modify SMPL_MODEL_ROOT in PATHS.yml accordingly.
This demo also requires the object templates from BEHAVE dataset. Download the object templates from here and modify BEHAVE_PATH in file PATHS.yml accordingly.
Run demo example with:
python demo.py chore-release -s example -on basketball results will be saved to example/000000117377/demo.
Please follow the instructions here to download the BEHAVE dataset.
After download and unzip, modify the BEHAVE_PATH in file PATHS.yml accordingly.
Specify the root path PROCESSED_PATH in file PATHS.yml to where you want the processed data to be saved.
Preprocess one sequence:
python preprocess/preprocess_scale.py -s [path to one sequence]Alternatively you can run python preprocess/preprocess_scale.py -a to process all sequences sequentially.
Specify the number of GPUs by nproc_per_node and run the following to start training:
python -m torch.distributed.launch --nproc_per_node=4 --use_env train_launch.py -en chore-releaseFor BEHAVE dataset, we test on images from kinect one (k1) and evaluate on images where object is occluded less than 70%. To compute the occlusion ratio, you need to render full object masks.
Our rendered full object masks can be downloaded here. Download and extract them to the same path where BEHAVE sequences were extracted.
After downloading the pretrained model, you can test one sequence from behave data with:
python recon/recon_fit_behave.py chore-release --save_name chore-release -s [path to one sequence]To align reconstruction with input images for BEHAVE data, you will need to set crop_cent to False when using this function.
Run the following to compute errors reported in table 1:
python recon/evaluate.py If you use our code, please cite:
@inproceedings{xie2022chore,
title = {CHORE: Contact, Human and Object REconstruction from a single RGB image},
author = {Xie, Xianghui and Bhatnagar, Bharat Lal and Pons-Moll, Gerard},
booktitle = {European Conference on Computer Vision ({ECCV})},
month = {October},
organization = {{Springer}},
year = {2022},
}If you use BEHAVE dataset, please also cite:
@inproceedings{bhatnagar22behave,
title = {BEHAVE: Dataset and Method for Tracking Human Object Interactions},
author={Bhatnagar, Bharat Lal and Xie, Xianghui and Petrov, Ilya and Sminchisescu, Cristian and Theobalt, Christian and Pons-Moll, Gerard},
booktitle = {{IEEE} Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {jun},
organization = {{IEEE}},
year = {2022},
}Copyright (c) 2022 Xianghui Xie, Max-Planck-Gesellschaft
Please read carefully the following terms and conditions and any accompanying documentation before you download and/or use this software and associated documentation files (the "Software").
The authors hereby grant you a non-exclusive, non-transferable, free of charge right to copy, modify, merge, publish, distribute, and sublicense the Software for the sole purpose of performing non-commercial scientific research, non-commercial education, or non-commercial artistic projects.
Any other use, in particular any use for commercial purposes, is prohibited. This includes, without limitation, incorporation in a commercial product, use in a commercial service, or production of other artefacts for commercial purposes.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
You understand and agree that the authors are under no obligation to provide either maintenance services, update services, notices of latent defects, or corrections of defects with regard to the Software. The authors nevertheless reserve the right to update, modify, or discontinue the Software at any time.
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software. You agree to cite the CHORE: Contact, Human and Object REconstruction from a single RGB image paper in documents and papers that report on research using this Software.