Project page: https://david-svitov.github.io/HAHA_project_page/
Arxiv paper: https://arxiv.org/pdf/2404.01053.pdf
YouTube video: https://youtu.be/vBzdAOKi1w0

We present HAHA - a novel approach for animatable human avatar generation from monocular input videos. The proposed method relies on learning the trade-off between the use of Gaussian splatting and a textured mesh for efficient and high fidelity rendering. We demonstrate its efficiency to animate and render full-body human avatars controlled via the SMPL-X parametric model. Our model learns to apply Gaussian splatting only in areas of the SMPL-X mesh where it is necessary, like hair and out-of-mesh clothing. This results in a minimal number of Gaussians being used to represent the full avatar, and reduced rendering artifacts. This allows us to handle the animation of small body parts such as fingers that are traditionally disregarded. We demonstrate the effectiveness of our approach on two open datasets: SnapshotPeople and X-Humans. Our method demonstrates on par reconstruction quality to the state-of-the-art on SnapshotPeople, while using less than a third of Gaussians. HAHA outperforms previous state-of-the-art on novel poses from X-Humans both quantitatively and qualitatively.

git clone --recursive https://github.com/david-svitov/HAHA.git

The easiest way to build an environment for this repository is to use docker image. To build it, make the following steps:

1. Build the image with the following command:

bash docker/build.sh

2. Start a container:

bash docker/run.sh

It mounts root directory of the host system to /mounted/ inside docker and sets cloned repository path as a starting directory.

3. Inside the container install diff_gaussian_rasterization-alphadep. (Unfortunately, docker fails to install it during image building)

bash gaussian_install.sh

models/smplx
    ├── SMPLX_FEMALE.pkl
    ├── SMPLX_FEMALE.npz
    ├── SMPLX_MALE.pkl
    ├── SMPLX_MALE.npz
    ├── SMPLX_NEUTRAL.pkl
    ├── SMPLX_NEUTRAL.npz

white - Results rendered with white background
black - Results rendered with black background

Or Download all checkpoints and renders.

For X-Humans all renders are provided in novel poses.

Or Download all checkpoints and renders.

To generate an animation with an avatar, use the following command:

python main.py --base=./configs/<config name>.yaml --pretrained=/<checkpoints>/OPTIMIZE_OPACITY_10500.ckpt --test_mode

To train an avatar, run the code as follows. Make sure to modify all necessary paths in the config file.

python main.py --base=./configs/<config name>.yaml

If you find this code useful for your research, please consider citing:

@misc{svitov2024haha,
      title={HAHA: Highly Articulated Gaussian Human Avatars with Textured Mesh Prior}, 
      author={David Svitov and Pietro Morerio and Lourdes Agapito and Alessio Del Bue},
      year={2024},
      eprint={2404.01053},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}