First-order-motion-model

Implementation of First order motion model for Image Animation paper

Installation

To run this repo, you require atleast python>=3.10,<=3.12. We made use of python==3.11.5.

Create a new virtual environment and run the following command to install the dependenices:

pip install -r requirements.txt

Single GPU Training

Run the following command to start training:

python train.py --data_path /path/to/data --config_path /configs/*.yaml --log_path /path/to/logs

Multi GPU Training

Move to "multi_gpu" branch of this repo and run the following:

python multi_gpu_train.py --data_path /path/to/data --config_path /configs/*.yaml --log_path /path/to/logs

Run Inference using trained model

You can run the following command to animate any object using the trained model:

python run_inference.py --source_path source.gif --driving_path driving.gif --config_path /configs/*.yaml --checkpoint_path /path/to/ckp

Results

I've been able to train the model on the MGIF dataset. You can download the processed version here. The following is the result that I've obtained,

Citation

@InProceedings{Siarohin_2019_NeurIPS,
  author={Siarohin, Aliaksandr and Lathuilière, Stéphane and Tulyakov, Sergey and Ricci, Elisa and Sebe, Nicu},
  title={First Order Motion Model for Image Animation},
  booktitle = {Conference on Neural Information Processing Systems (NeurIPS)},
  month = {December},
  year = {2019}
}