ScoreHOI: Physically Plausible Reconstruction of Human-Object Interaction via Score-Guided Diffusion (ICCV 2025)
The repository contains the official implementation for the paper "ScoreHOI: Physically Plausible Reconstruction of Human-Object Interaction via Score-Guided Diffusion" (ICCV 2025).
ScoreHOI is an innovative framework for human-object interaction reconstruction that effectively incorporates physical constraints during the sampling process, providing physically plausible reconstruction results.
We recommend you to use an Anaconda virtual environment. If you have installed Anaconda, run the following commands to create and activate a virtual environment.
conda create -n scorehoi python=3.9
conda activate scorehoi
# Install PyTorch
pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118
# Other requirements
pip install -r requirements.txtWe follow the same data preparation steps as CONTHO_Data.
For the checkpoint and the features extracted from the PointNeXt, please refer to our huggingface repository.
data format:
|-- checkpoint
| |-- backbone.pth.tar
| |-- scorehoi.pth.tar
|-- data
| |-- base_data
| | |-- annotations
| | |-- backbone_models
| | |-- human_models
| | | |-- smplh
| | |-- object_models
| | | |-- behave
| | | | |-- _info_feature.pkl
| | | |-- intercap
| | | | |-- _info_feature.pkl
| |-- BEHAVE
| | |-- dataset.py
| | |-- sequences
| | | |-- Date01_Sub01_backpack_back
| | | |-- Date01_Sub01_backpack_hand
| | | |-- ...
| | | |-- Date07_Sub08_yogamat
| |-- InterCap
| | |-- dataset.py
| | |-- sequences
| | | |-- 01
| | | |-- 02
| | | |-- ...
| | | |-- 10
torchrun --master_port 29591 --nproc_per_node 8 \
main/train_diffusion.py \
--gpu 0,1,2,3,4,5,6,7 \
--checkpoint checkpoints/backbone.pth.tar \
--exp train_scorehoi \
--diffusion_dir checkpoints/scorehoi.pth.tar \
--distributed \
--not_continue
We would like to express our sincere thanks to CONTHO, for our code is partially borrowing from them.
We also thank CHORE, InterCap for their high quailty datasets.
- If you find an error as below, please refer to this link for help.
RuntimeError: Subtraction, the `-` operator, with a bool tensor is not supported. If you are trying to invert a mask, use the `~` or `logical_not()` operator instead.@article{li2025scorehoi,
title={ScoreHOI: Physically Plausible Reconstruction of Human-Object Interaction via Score-Guided Diffusion},
author={Li, Ao and Liu, Jinpeng and Zhu, Yixuan and Tang, Yansong},
journal={arXiv preprint arXiv:2509.07920},
year={2025}
}
This code is distributed under an MIT LICENSE.