pixel
A PyTorch-based Generative Adversarial Network (GAN) for training and generating pixel art images.
Core trick
A DCGAN generator and discriminator trained on a custom pixel-art dataset, with a Tkinter GUI sampling new images from saved weights.
Tags
Description
A PyTorch-based Generative Adversarial Network (GAN) for training and generating pixel art images.
README excerpt
# Pixel A PyTorch-based Generative Adversarial Network (GAN) for training and generating pixel art images. ## Table of Contents - [Overview](#overview) - [Project Structure](#project-structure) - [Architecture](#architecture) - [Workflow](#workflow) - [Installation](#installation) - [User Guide](#user-guide) - [Configuration](#configuration) - [Examples](#examples) --- ## Overview This project implements a Deep Convolutional GAN (DCGAN) to generate pixel art images. It includes: - **Training Pipeline**: Train a GAN model on your custom image dataset - **Image Generation**: Generate new images using trained models - **Interactive GUI**: Tkinter-based interface for real-time generation - **GGUF Support**: Convert and use GGUF quantized models --- ## Project Structure ``` pixel-model-trainer/ │ ├── trainer.py # GAN training script ├── generator.py # Image generation (GUI + CLI) │ ├── data/ # Training data │ ├── attributes.csv # Dataset metadata │ └── images/ # Training images (punk000.png, punk001.png, ...) │ ├── models/ # Trained model storage │ └── generator_model.safetensors │ ├── generated/ # Generated image outputs │ ├── output.png # Grid visualizations │ └── individual/ # Individual generated images │ ├── gen_images/ # Training progress images │ ├── epoch_0.png │ ├── epoch_1.png │ └── ... │ └─