Skip to content
Projects

zhengyuf/PointAvatar has 342 GitHub stars and 24 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-19 03:02 UTC. Last push 2024-04-05.

How these counts are measured

Measured star growth

in 24h
0
in 7 days
Collecting history
in 15 days
Collecting history
in 1 month
Collecting history

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

Official Repository for CVPR 2023 paper PointAvatar: Deformable Point-based Head Avatars from Videos.

Creative & Media

README.md

Original

PointAvatar: Deformable Point-based Head Avatars from Videos

Paper | Video Youtube | Project Page

Official Repository for CVPR 2023 paper PointAvatar: Deformable Point-based Head Avatars from Videos.

Getting Started

  • Clone this repo: git clone [email protected]:zhengyuf/pointavatar.git
  • Create a conda or python environment and activate. For e.g., conda create -n point-avatar python=3.9; conda activate point-avatar.
  • Install PyTorch 1.11.0 with conda or pip (instructions). This version works with both PyTorch3d and functorch.
  • We made small modifications to PyTorch3d (0.6.2), so clone our version and install:
git clone [email protected]:zhengyuf/pytorch3d.git
cd pytorch3d
git checkout point-avatar
pip install -e .
  • Install other requirements: cd ../pointavatar; pip install -r requirement.txt
  • Download FLAME model, choose FLAME 2020 and unzip it, copy 'generic_model.pkl' into ./code/flame/FLAME2020

Preparing dataset

Our data format is the same as IMavatar. You can download a preprocessed dataset from the ETH Zurich server (subject 1, subject 2 and subject 3). You can run download_data.bash to download both datasets and pre-trained models.

If you'd like to generate your own dataset, please follow the instructions in the IMavatar repo.

Link the dataset folder to ./data/datasets. Link the experiment output folder to ./data/experiments.

Pre-trained model

Download a pretrained model from ETH Zurich server (subject 1, subject 2 and subject 3). See download_data.bash. Uncompress and put into the experiment folder ./data/experiments.

Training

python scripts/exp_runner.py --conf ./confs/subject1.conf [--is_continue]

Evaluation

Set the is_eval flag for evaluation, optionally set checkpoint (if not, the latest checkpoint will be used) and load_path

python scripts/exp_runner.py --conf ./confs/subject1.conf --is_eval [--checkpoint 60] [--load_path ...]

GPU requirement

We train our models with a single Nvidia 80GB A100 GPU. It's also possible to train PointAvatar with less CUDA memory by limiting the number of points. For e.g., with a 24GB Quadro RTX 6000, you can set train.max_points_training and point_cloud.max_points to 240000. For evaluation, by default, the latest checkpoint is used. To reduce memory usage, you can use earlier checkpoints by adding —-checkpoint 60 (or 55).

Citation

If you find our code or paper useful, please cite as:

@inproceedings{Zheng2023pointavatar,
  author    = {Yufeng Zheng and Wang Yifan and Gordon Wetzstein and Michael J. Black and Otmar Hilliges},
  title     = {PointAvatar: Deformable Point-based Head Avatars from Videos},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)}, 
  year = {2023}
}