High-Fidelity Face Age Transformation via Hierarchical Encoding and Contrastive Learning (HECL)

This README provides instructions for setting up, training, and testing the High-Fidelity Face Age Transformation via Hierarchical Encoding and Contrastive Learning (HECL), a state-of-the-art model for age transformation in images. The guide assumes operation on a Linux environment and recommends the use of a GPU for execution.

Environment Setup, Training, and Inference Instructions

Environment Setup

1. Create a Conda Environment: Use the environment.yml file provided in the repository to create a conda environment by running the following command:

   conda env create -f environment.yml
   

Data Preparation

1. Download Datasets: Ensure to download the following datasets and place them in the directory where HECL is located(not inside the HECL directory):

   • FFHQ-Aging-Dataset: GitHub Link
   • Cross-Age-Face Dataset: GitHub Link
   • All-Age-Faces-Dataset: Google Drive Link

   Unzip All-Ages-Faces-Dataset/results/cropped_imgs.zip.

2. Preprocess the Datasets: Execute the Python scripts below to preprocess the datasets. This step is crucial for preparing the data for training:

   python3 datasets/create_dataset.py
   python3 datasets/create_dataset_caf.py
   python3 datasets/create_dataset_allagesdataset.py

   Alternatively, you can use the provided shell script to automate this process:

   sh create_dataset.sh

Training

First, turn on visdom on another terminal. This will open visdom server on localhost:8097 by default.

visdom

To train the model, modify the --dataroot, --name parameters in train.sh script according to your needs, then run:

sh train.sh

Pretrained Models

Download checkpoints.zip Google Drive Link, unzip and put the checkpoints folder in the HECL directory. The inference can be done using pretrained models.

Inference

For inference, modify the --dataroot, --name, --which_epoch, and --checkpoint_dir parameters in test.sh script according to your needs, then run:

sh test.sh

Notes

Ensure that you have correctly placed the datasets in the required directory before preprocessing.

Modify the script parameters carefully to match your local setup for successful training and testing.