title	emoji	colorFrom	colorTo	sdk	sdk_version	app_file	pinned	license
Fashion-MNIST DDPM Generator	👗	purple	pink	gradio	4.0.0	app.py	false	mit

Fashion-MNIST DDPM Generator

Generate realistic fashion items using a trained Denoising Diffusion Probabilistic Model (DDPM).

Features

• 10 Fashion Categories: T-shirt/top, Trouser, Pullover, Dress, Coat, Sandal, Shirt, Sneaker, Bag, Ankle boot
• Classifier-Free Guidance: Control the strength of class conditioning
• Batch Generation: Generate multiple samples at once
• All Categories View: Generate samples for all 10 categories simultaneously

Model Details

This app uses a DDPM trained on the Fashion-MNIST dataset with:

• U-Net architecture with attention
• Cosine noise schedule
• Classifier-free guidance for conditional generation
• EMA (Exponential Moving Average) weights for better quality

Usage

1. Select a Category: Click on any fashion category button to generate samples
2. Adjust Settings:
   • Number of Samples: How many images to generate (1-16)
   • Guidance Scale: Higher values = stronger class conditioning (1.0-10.0)
3. Generate All: Use the "Generate All Categories" tab to see samples from all categories

Technical Details

• Model: Small U-Net (32 base channels)
• Training: 150 epochs on Fashion-MNIST
• Diffusion Steps: 500
• Image Size: 28x28 (upscaled to 224x224 for display)

Credits

Built with:

• PyTorch
• Gradio
• Fashion-MNIST Dataset