EnrichGAN: Exploiting Enriched Discriminator Representations for Training GANs under Limited Data

This repository contains the official PyTorch implementation of the paper "EnrichGAN: Exploiting Enriched Discriminator Representations for Training GANs under Limited Data".

1. Repository Structure

• models.py: Contains the architectural definitions of the generator, discriminator, and decoder.
• models_SVA.py: Contains the definition of the SVA module.
• operation.py: Provides helper functions and data-loading pipelines.
• train.py: The main training script. Intermediate results and model checkpoints are automatically saved to the train_results/ directory during execution.
• eval.py: Synthesizes images using a pre-trained generator and saves them to a designated directory for subsequent quantitative evaluation.
• benchmarking/: Scripts for computing quantitative evaluation metrics. (Note: The official PyTorch Inception model is downloaded automatically upon first use).
• lpips/: Implementation of the perceptual loss. (The corresponding pre-trained network is downloaded automatically).

2. Usage

Training

To train the model, place all your training images into a single directory and run the following command:

python train.py --path /path/to/RGB-image-folder

Generation

Upon completing the training process, you can synthesize image samples using the saved checkpoint by running:

cd ./train_results/name_of_your_training/
python eval.py --n_sample 5000 

Evaluation

After generating the images, you can compute the evaluation metrics via:

cd ./benchmarking/
python metrics_compute.py --path_a /path/to/RGB-image-folder --path_b ./train_results/name_of_your_training