HEAR-Net Pytorch Implementation

This repository contains an unofficial implementation of HEAR-Net(Heuristic Error Acknowledging Refinement Network) introduced in the following paper:

Li, Lingzhi, et al. "Faceshifter: Towards high fidelity and occlusion aware face swapping." arXiv preprint arXiv:1912.13457 (2019).

Environment

Dockerfile will be updated.

Usage

Train

Please run the following command after placing all the needed datasets:

python3 train.py --source_images {SOURCE_PATH} --target_images {TARGET_PATH} --swapped_images {SWAPPED_PATH} heuristic_errors {ERROR_PATH}

Inference

Please run the following command after placing all the needed datasets:

python3 train.py --swapped_images {SWAPPED_PATH} heuristic_errors {ERROR_PATH}