FMA-3D

A method for adding a mask on a non-masked face image. Given a real masked face image (a) and a non-masked face image (d), we synthesize a photo-realistic masked face image with the mask from (a) and the facial area from (d).

Some Results by FMA-3D

Requirements

• python >= 3.7.1
• pytorch >= 1.1.0

Usage

• Extract the landmarks. You can extract the 106 landmarks by our face sdk or any other methods.
• Add a mask on a face image. You can refer to add_mask_one.py as an example.

python add_mask_one.py

Speed Up

Some advice:

• Write the whole process by multi-processing.
• Write the function of render in face_masker.py by c++.

Reference

This project is mainly inspired by PRNet.