Generative Adversarial Networks-for-Biomaterial-Discovery

GANs for Image Generation using Biomaterial Topography Dataset

This project contains the code to train a suite of GAN Variants for Biomaterial Discovery. It includes a PyTorch implementation of Generative Adversarial Networks with several building blocks used in the model architecture and scripts for training and logging.

Project Structure

• GANs - Generated Topographic Images/: In this file, GANs-generated Images can be found.

• Launch/: This file contains bash scripts where we can parse the custom training options.

• models/:

  • acgan.py: Implementation of Auxiliary Classifier GAN (ACGAN) model with convo layers.
  • acvanilla.py: Implementation of the vanilla ACGAN model.
  • discriminators.py: Implementation of various discriminator networks used in the GAN architecture.
  • generators.py: Implementation of various generator networks used in the GAN architecture.
  • models.py: Implementation of Base Model (parent) definitions and configurations for the GAN architecture.
  • networks.py: Implementation of Base Network (parent) definitions and configurations for the GAN architecture.
  • vanillagan.py: Implementation of vanilla GAN model.
  • wgan.py: Implementation of Wasserstein GAN (WGAN) and WGAN-GP model.

• options/:

  • base_options.py: Basic Command-line arguments for the training script.
  • train_options.py: Hyperparameter Command-line arguments for the training script.

• utils/:

  • images_utils.py: Utilities for image handling.
  • custom_layers.py: This file contains scripts for custom layers used in the GAN architecture.
  • losses.py: This file contains scripts for loss functions used in the GAN training process.
  • tb_visualizer.py: This file provides scripts for a TensorBoard visualizer for monitoring GAN training progress.
  • utils.py: This file contains scripts for various utility functions used in the GAN project.
  • weights_init.py: This file contains scripts for weight initialization functions for the GAN architecture.

• call_methods.py: This file contains scripts for dynamically creating models, networks, datasets, and data loaders based on provided names and options.

• train.py: Script for training the model with TensorBoard logging.

• predict.py: Script for generating the biomaterial images without "No-Grad".

• evalaute.py: Scripts for evaluating the GAN-generated images using FID Scores.

Requirements

To run the code, you need the following:

• Python 3.8 or above
• PyTorch 1.7 or above
• torchvision
• tqdm
• matplotlib
• TensorboardX 2.7.0

Install the necessary packages using pip:

Dataset

The training scripts are set up to use the Biomaterial dataset with 2176 Samples, which are loaded from the local machine. If you wish to use a different dataset, you'll need to modify the images_utils.py file and potentially the training scripts to handle your dataset's loading and processing.