language:
- en
license: mit
metrics:
- recall
- precision
- f1
pipeline_tag: image-to-3d
tags:
- 3d-human-reconstruction
- human-scene-contact
- monocular-rgb
- mesh-reconstruction
- pose-aware
- icme-2026
GraphiContact: Pose-aware Human-Scene Robust Contact Perception for Interactive Systems
This repository contains the pre-trained checkpoints for GraphiContact, a novel framework for monocular vertex-level human-scene contact prediction and 3D human mesh reconstruction.
Paper (arXiv) | Official GitHub Repository
Overview
GraphiContact jointly addresses vertex-level contact prediction and single-image 3D human mesh reconstruction. It uses reconstructed body geometry as a scaffold for contact reasoning, integrating pose-aware features with human-scene interaction understanding.
Key Features
- Pose-aware Framework: Transfers complementary human priors from pretrained Transformer encoders to predict per-vertex human-scene contact on the reconstructed mesh.
- SIMU Training Strategy: Introduces a Single-Image Multi-Infer Uncertainty (SIMU) training strategy with token-level adaptive routing. This simulates occlusion and noisy observations during training while preserving efficient single-branch inference at test time.
- Robust Perception: Specifically designed to handle real-world scenarios with perceptual noise and occlusions, making it suitable for interactive systems like embodied AI and rehabilitation analysis.
Installation and Usage
For detailed instructions on environment setup, downloading model weights, and running inference demos, please refer to the official GitHub repository.
Citation
If you find this work useful for your research, please consider citing the paper:
@inproceedings{lin2026graphicontact,
title={GraphiContact: Pose-aware Human-Scene Robust Contact Perception for Interactive Systems},
author={Lin, Aveiro and others},
booktitle={IEEE International Conference on Multimedia and Expo (ICME)},
year={2026}
}
License
The research code is released under the MIT license. Note that the model has dependencies on the SMPL and MANO models, which are subject to their own Software Copyright License for non-commercial scientific research purposes.