Doom Image Diffusion Project

This project implements a diffusion model for generating and processing images, inspired by the game Doom. It includes a dataset class for handling image data, a U-Net-based neural network for image generation, and a noise schedule for diffusion.

Features

• Dataset Handling: The DoomImages class processes image datasets, ensuring proper resizing, normalization, and loading.
• Diffusion Model: Implements a U-Net architecture with sinusoidal timestep embeddings for image generation.
• Noise Schedule: Precomputes a linear noise schedule for diffusion steps.

Project Structure

• image_dataset.py: Contains the DoomImages dataset class for loading and preprocessing images.
• model.py: Defines the U-Net model, sinusoidal timestep embeddings, and noise schedule.
• .gitignore: Specifies files and directories to exclude from version control.
• README.md: Documentation for the project.

Requirements

• Python 3.12+
• PyTorch
• NumPy
• Pillow

Installation

1. Clone the repository:

   git clone <repository-url>
   cd <repository-folder>

Usage

Dataset Preparation

Place your .png images in a folder (e.g., data/). The DoomImages class will automatically load and preprocess them.

Training the Model

1. Import the dataset and model:

   from image_dataset import DoomImages
   from libs.models.simple_unet import SimpleUNet

2. Initialize the dataset and model:

   dataset = DoomImages(folder='data/', image_size=64)
   model = SimpleUNet()

3. Train the model using your preferred training loop.

License

This project is licensed under the MIT License. See the LICENSE file for details.