GenZI: Zero-Shot 3D Human-Scene Interaction Generation

By Lei Li and Angela Dai. (CVPR 2024)

Can we synthesize 3D humans interacting with scenes without learning from any 3D human-scene interaction data? We propose GenZI, the first zero-shot approach to generating 3D human-scene interactions. Key to GenZI is our distillation of interaction priors from large vision-language models (VLMs), which have learned a rich semantic space of 2D human-scene compositions.

Given a natural language description and a coarse point location of the desired interaction in a 3D scene, we first leverage VLMs to imagine plausible 2D human interactions inpainted into multiple rendered views of the scene. We then formulate a robust iterative optimization to synthesize the pose and shape of a 3D human model in the scene, guided by consistency with the 2D interaction hypotheses.

In contrast to existing learning-based approaches, GenZI circumvents the conventional need for captured 3D interaction data, and allows for flexible control of the 3D interaction synthesis with easy-to-use text prompts. Extensive experiments show that our zero-shot approach has high flexibility and generality, making it applicable to diverse scene types, including both indoor and outdoor environments.

Citation

If you find our work useful in your research, please consider citing:

@InProceedings{li2024genzi,
    title     = {{GenZI}: Zero-Shot {3D} Human-Scene Interaction Generation},
    author    = {Li, Lei and Dai, Angela},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2024}
}

Install

# Create a new conda environment
conda create -n genzi python=3.8
conda activate genzi

# Install PyTorch: assume cuda 11.7
conda install -y pytorch=2.0.1 torchvision=0.15.2 torchaudio=2.0.2 pytorch-cuda=11.7 -c pytorch -c nvidia

# Install other libraries
pip install -r requirements.txt
pip install -U --no-deps smplx==0.1.28 git+https://github.com/nghorbani/human_body_prior.git
conda install -y open3d-admin::open3d=0.10.0.0
# Install PyTorch3D: https://github.com/facebookresearch/pytorch3d/blob/main/INSTALL.md
# Install AlphaPose: https://github.com/MVIG-SJTU/AlphaPose (try Cython==0.29.35 setuptools==65.7.0 if anything complains)
# Install torch-mesh-isect: https://github.com/vchoutas/torch-mesh-isect
# (Try numpy==1.23.4 if numpy complains)

Data

The 3D scene data and our generation results can be found here. Extract the content of the zipped file to the root directory of the code.

Running

# See more running settings therein
bash run.sh

References

Diffusers by Hugging Face.
Fang et al. AlphaPose: Whole-Body Regional Multi-Person Pose Estimation and Tracking in Real-Time.
Pavlakos et al. Expressive Body Capture: 3D Hands, Face, and Body from a Single Image.
Hassan et al. Resolving 3D Human Pose Ambiguities with 3D Scene Constraints.
Zhao et al. Compositional Human-Scene Interaction Synthesis with Semantic Control.

This work is licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International.

craigleili/GenZI