Landmarkgan: Synthesizing faces from landmarks

This repository is the implementation of our work, Landmarkgan: Synthesizing faces from landmarks

Preparation

1.landmark coordinates order:
    facial boundary landmarks are in the last.
    
2.pip install:
    torch(pytorch 1.0.1. Other versions might also work.)
    easydict
    dsntnn
    pyyaml
    opencv-python
    numpy
    pickle

Use opencv-python 3.x and ignore opencv version warning.

Generation

Step 1

• download lm_AE checkpoints from here.
• unzip it and put all files in ./model/ckpt/MODEL_0301_1_patchgan_lmconsis_lm_AE/
• Modify the absolute path in 00_FSNet_lm_AE.yml(line 20).

Step 2

• download tl2f checkpoints from here.
• unzip it and put all files in ./model/ckpt/MODEL_0130_1_patchgan_lmconsis_one_decoder/
• Modify the absolute path in 00_FSNet_onedecoder.yml(line 20).

Step 3

• python 0830_tkdemo.py
• demo

Bibtex

@article{sun2022landmarkgan,
  title={Landmarkgan: Synthesizing faces from landmarks},
  author={Sun, Pu and Li, Yuezun and Qi, Honggang and Lyu, Siwei},
  journal={Pattern Recognition Letters},
  volume={161},
  pages={90--98},
  year={2022},
  publisher={Elsevier}
}