CycleGAN for Image To Image Translation - Wildlife

Overview

Imagine receiving a self-portrait from the renowned Dutch Post-Impressionist painter Vincent van Gogh. Seems impossible, right? In recent years, Microsoft researchers have developed the CycleGAN model, specifically designed for unpaired image-to-image translation. This powerful model has a wide range of applications, from enhancing photographs to transforming seasons or styles. With CycleGAN, you can translate any image to another domain, regardless of whether they are paired.

---

Aim

To build the CycleGAN Model in PyTorch for translating animal image to other domain irrespective of the pairing between two images.

---

Data Description

The dataset includes images of horses in domain A and zebras in domain B, divided into train and test sets.

---

Tech Stack

• Language: Python
• Libraries: torch, torchvision, numpy, pillow

---

Approach

1. Introduction to CycleGAN
2. Introduction to ResNet
3. Loss function in CycleGAN
4. Building Model on PyTorch
5. Model training

---

Modular Code Overview

1. data: Contains the horse2zebra dataset, split into train and test sets with domain A as Horse and domain B as Zebra.
2. images: Contains translated images from domain A to B.
3. saved_model: Placeholder for saving trained models.
4. cyclegan.py: The main script for training the CycleGAN model.
5. datasets.py: Contains code for data loading and preprocessing.
6. models.py: Defines the CycleGAN model architecture.
7. utils.py: Utility functions used in training.
8. requirements.txt: Lists all the required libraries with their respective versions. Install them using pip install -r requirements.txt.

---