Awesome

[ECCV 2022] Motion Transformer for Unsupervised Image Animation

This is the project page of the paper Motion Transformer for Unsupervised Image Animation (ECCV 2022).

2022.8.15: Codes are released.

Environments

The model is trained on 8 Tesla V100 cards, PyTorch version 1.6 and 1.8 with Python 3.6 are tested fine. Basic installations are given in requiremetns.txt.

pip install -r requirements.txt

Datasets

Following FOMM to download TaiChiHD,Voxceleb1, and MGIF. And for the TED384, following MRAA. After downloading and pre-processing, the dataset should be placed in the ./data folder or you can change the parameter root_dir in the yaml config file. Note that we save the video dataset in PNG frames format for better training IO performance. All train and test video frames are specified in txt files in the ./data folder. The tree structure of the dataset path is given in the following.

|-- data/taichi-png, data/TED384-png, data/voxceleb1-png, data/moving-gif-png
    |-- train
        |-- video1
            |-- 00000.png
            |-- 00001.png
            |-- ...
        |-- video2
            |-- 00000.png
            |-- 00001.png
            |-- ...
        |-- ...
    |-- test
        |-- video1
            |-- 00000.png
            |-- 00001.png
            |-- ...
        |-- video2
            |-- 00000.png
            |-- 00001.png
            |-- ...
        |-- ...

Checkpoints

Google drive Baiduyun passwd:qp6v

Training

We train the model on 8 cards and use pytorch DistributedDataPrallel.

CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --nproc_per_node=8 run.py --config config/dataset.yaml

Evaluation

Evaluate video reconstruction with the following command, for more metrics, we recommend seeing FOMM-Pose-Evaluation.

CUDA_VISIBLE_DEVICES=0 python run.py --mode reconstruction --config path/to/config --checkpoint path/to/model.pth  

Demo

To make a demo animation, specify the driving video and source image, the result video will be saved to result.mp4.

python demo.py --config path/to/config --checkpoint path/to/model.pth --driving_video path/to/video.mp4 --source_image path/to/image.png --result_video path/to/result.mp4 --adapt_scale

Animation

Citation

@inproceedings{tao2022motion,
title={Motion Transformer for Unsupervised Image Animation},
author={Tao, Jiale and Wang, Biao and Ge, Tiezheng and Jiang, Yuning and Li, Wen and Duan, Lixin},
booktitle={European Conference on Computer Vision},
year={2022}
}

Acknowledgements

The implementation is partially borrowed from FOMM, TokenPose and TransPose, we thank the authors for their excellent works.