README.md

December 10, 2025 · View on GitHub

FPEM: Face Prior Enhanced Facial Attractiveness Prediction for Live Videos with Face Retouching [ICCV 2025]

Hui Li, Xiaoyu Ren, Hongjiu Yu, Ying Chen*, Kai Li, L Wang, Xiongkuo Min, Huiyu Duan, Guangtao Zhai, Xu Liu

News

  • [2025-11-03] The FPEM model is released.
  • [2025-11-03] The paper is released on ICCV2025🔥.
  • [2025-12-10] The LiveBeauty dataset is released.

TODO

  • Release the FPEM model.
  • Release the dataset.

Dataset Statistics

  • Examples of the face images in our LiveBeauty dataset.

We provide two version of LiveBeauty dataset, you can download them from here. One version is the same dataset as the one in the paper denoted as LiveBeauty-essay, the other is a complete-face-version denoted as LiveBeauty-public. The complete-face-version contains complete face images and the corresponding original frames of the live videos.

To avoid disputes, the celebrity faces are removed fom the dataset.

Experimental Results

  • Experimental results of various SOTA methods and our FPEM across three FAP datasets.

The pretrained checkpoints for FPEM described in the paper are available in the FPEM-release/pretrained folder.

Installation

# Create your environment
conda create -n fpem python=3.8
conda activate fpem
pip install --upgrade pip

# Install pytorch with cuda
pip install torch==1.13.1+cu117 torchvision==0.14.1+cu117 torchaudio==0.13.1 --extra-index-url https://download.pytorch.org/whl/cu117

# Install Dependencies
pip install -r requirements.txt

There is no need to install all dependencies listed in the requirements.txt, but you should make sure protobuf==3.19.0

Train

chmod +x train.sh
# run train.sh for <base> experiment, you can creat your own exp name which must have config.yml in it
bash train.sh experiments/base

Test

chmod +x test.sh
# replace your ckpt.pth path in config.yml-> saver -> resume_model 
bash test.sh experiments/base

Citation

If you find this work useful for your research, please consider citing our paper:
@inproceedings{li2025fpem,
  title={FPEM: Face Prior Enhanced Facial Attractiveness Prediction for Live Videos with Face Retouching},
  author={Hui Li, Xiaoyu Ren, Hongjiu Yu, Ying Chen*, Kai Li, L Wang, Xiongkuo Min, Huiyu Duan, Guangtao Zhai, Xu Liu},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision, ICCV 2025},
  year={2025}
}