by dawei6875797

219 Stars 92 Forks Last release: Not found 18 Commits 0 Releases

Available items

No Items, yet!

The developer of this repository has not created any items for sale yet. Need a bug fixed? Help with integration? A different license? Create a request here:

Face Aging with Identity-Preserved Conditional Generative Adversarial Networks

This repo is the official open source of Face Aging with Identity-Preserved Conditional Generative Adversarial Networks, CVPR 2018 by Zongwei Wang, Xu Tang, Weixin Luo and Shenghua Gao. It is implemented in tensorflow. Please follow the instructions to run the code. scalars_framework

1. Installation

  • Install 3rd-package dependencies of python (listed in requirements.txt)
pip install -r requirements.txt
  • Other libraries
    CUDA 8.0
    Cudnn 6.0
    ## 2. Download datasets We use the Cross-Age Celebrity Dataset for training and Evaluation. More details about this dataset, please refer to (http://bcsiriuschen.github.io/CARC/). After face detection, aligning and center cropping, we split images into 5 age groups: 11-20, 21-30, 31-40, 41-50 and 50+.

3. Test on saved models

Download the trained face aging model(https://1drv.ms/u/s!AlUWwwOcwDWobCqmuFyKGIt4qaA) and place models files in checkpoints/0conv5lsgantransferg750.5f-4a30. * Test images are in images/test, and some training images that belong to 11-20 age group are in images/train.

* Running the sript to get aged faces
    python test.py

4. Train from scratch

  • Firstly, download the pre-trained alexnet model(https://1drv.ms/u/s!AlUWwwOcwDWobkptownyu5fjlfU) and age classfication model(https://1drv.ms/f/s!AlUWwwOcwDWocX-Z0IJftVbcoQ). Then unzip these files and place model files in checkpoints/pretrained.
    python age_lsgan_transfer.py \
    --gan_loss_weight=75 \
    --fea_loss_weight=0.5e-4 \
    --age_loss_weight=30 \
    --fea_layer_name=conv5 \
    --checkpoint_dir=./checkpoints/age/0_conv5_lsgan_transfer_g75_0.5f-4_a30 \
    sh age_lsgan_transfer.py
  • You can change the hyperparameters to adapt to your own dataset.

5. Experiment results

  • The aging effect of different methods. Within each dot box, the first row is our result.scalars_method_comparison
  • The aging effect of different age classification loss weights.scalars_age_loss_weight
  • The aging effect of different feature layer.scalars_layer


If you find this useful, please cite our work as follows:

    author={Z. Wang and X. Tang, W. Luo and S. Gao}, 
    booktitle={2018 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)}, 
    title={Face Aging with Identity-Preserved Conditional Generative Adversarial Networks}, 

We use cookies. If you continue to browse the site, you agree to the use of cookies. For more information on our use of cookies please see our Privacy Policy.