Gen AI - Assignment 03

This repository contains implementations of various Generative Adversarial Networks (GANs) as part of the Generative AI course assignment. The projects cover image generation, image-to-image translation, and domain translation.

Project Overview

The project is divided into three main parts:

1. DCGAN: Anime Face Generation (genai-A03_Q1.ipynb)

• Objective: Generate realistic anime faces from random noise.
• Model: Deep Convolutional Generative Adversarial Network (DCGAN).
• Dataset: Anime Faces dataset.
• Framework: PyTorch.

2. Pix2Pix: Sketch-to-Image & Colorization (Q2_A03_GEN_AI.ipynb)

• Objective: Convert black-and-white anime sketches into fully colored images.
• Model: Pix2Pix (Conditional GAN).
• Dataset: Anime sketch-to-photo pairs.
• Key Features: Uses a U-Net generator and a PatchGAN discriminator.

3. CycleGAN: Sketch to Photo Translation (Q3_A03_GEN_AI.ipynb)

• Objective: Unpaired image-to-image translation between the sketch domain and the photo domain.
• Model: Cycle-Consistent Adversarial Network (CycleGAN).
• Dataset: TU-Berlin sketches and custom photo collections.
• Key Features: Implements cycle-consistency loss and identity loss for stable translation without paired training data.

Technologies Used

• Python
• PyTorch
• Torchvision
• Matplotlib
• NumPy
• Kaggle T4 GPUs (for training)

Author

• Email: wareeshaashraf09@gmail.com

How to Use

1. Clone the repository:

   git clone https://github.com/wareeshayy/Gen_AI.git

2. Open the .ipynb files in Google Colab or Kaggle.
3. Ensure you have the required datasets or update the paths within the notebooks.
4. Run the cells to train or evaluate the models.