jnnan/chairs

Full-Body Articulated Human-Object Interaction

This is the code repository of Full-Body Articulated Human-Object Interaction - pdf - arxiv - project -

News

The metas.json containing the transformation matrix between each camera view and the world coordinate is uploade (which can make the human and the object on the ground).

Environment

We tested our code with CUDA 11.1, PyTorch 1.11.0, and torchvision 0.10.0.

After installing CUDA, PyTorch and torchvision, run the following command for installing other dependencies:

pip install -r requirements.txt

Data Preparation

Please download the data and weights from this link and extract the zip file to the root directory of this repository.

You need to download the SMPL-X models from here and unzip them to Data/body_models/smplx/.

Your directory should look like this:

├─ Data
    ├─ AHOI_ROOT
        ├─ Meshes_wt
        ├─ Metas
        ├─ object_part_voxel_64
    ├─ DATA_FOLDER
    ├─ IMG_FOLDER
        ├─ 0001
            ├─ 1
                ├─ 0000.png
                ├─ 0005.png
                ├─ 0010.png
                    ...
            ├─ 2
            ├─ 3
            ├─ 4
        ├─ 0002
            ...
    ├─ body_models
        ├─ smplx
    ├─ checkpoint
├─ ahoi_utils.py
├─ dataloaders.py
    ...