PyTorch Implementation of Image Animation with Perturbed Masks [CVPR 2022]

[Paper]

See samples and intermediate results in samples/main.html.

Method Overview

Installation

Tested using python 3.6.10.

You can install this repo as a Python package:

$ conda install pytorch==1.0.0 torchvision==0.2.1 cuda100 -c pytorch
$ pip install git+https://github.com/itsyoavshalev/Image-Animation-with-Perturbed-Masks.git

Datasets

In order to download the datasets, follow the instructions on https://github.com/AliaksandrSiarohin/first-order-model.

Training

To train using a specific dataset, run the following command:

$ python run.py --config config/dataset_name.yaml

Checkpoints and outputs will be saved into the log folder.

Reconstruction

In order to evaluate the video reconstruction using a specific dataset and a checkpoint path, run the following command:

$ python run.py --config config/dataset_name.yaml --mode reconstruction --checkpoint checkpoint_path

Outputs will be saved into the log folder.

Animation Demo

To run the image animation pipeline for a specific dataset and a checkpoint path, run the following command:

$ python run.py --config config/dataset_name.yaml --mode animate --checkpoint checkpoint_path

Outputs will be saved into the log folder.