🌀 Fashion-MNIST Diffusion (Class-Conditional UNet)
This repository hosts a class-conditional diffusion model trained on the Fashion-MNIST dataset.
The model uses a UNet-based architecture with timestep and label conditioning, implemented using PyTorch and integrated with the 🤗 Hugging Face Hub via PyTorchModelHubMixin.
🧩 Model Details
| Property | Description |
|---|---|
| Architecture | Conditional UNet with residual and skip connections |
| Conditioning | Class embedding (10 Fashion-MNIST labels) + timestep embedding |
| Framework | PyTorch |
| Pipeline | text-to-image (adapted for diffusion) |
| License | MIT |
| Author | Sherwin Roger |
The model learns to denoise images progressively, generating class-conditional samples (e.g., shoes, shirts, coats) from Gaussian noise through iterative diffusion steps.
🚀 Load from Hugging Face
You can directly load the model using huggingface_hub:
from huggingface_hub import hf_hub_download
import sys
import importlib.util
model_file = hf_hub_download(
repo_id="Sherwinroger002/fashion_mnist_diffusion_class_conditional",
filename="modeling.py"
)
spec = importlib.util.spec_from_file_location("modeling", model_file)
modeling = importlib.util.module_from_spec(spec)
sys.modules["modeling"] = modeling
spec.loader.exec_module(modeling)
model = modeling.EmbUnetModel.from_pretrained(
"Sherwinroger002/fashion_mnist_diffusion_class_conditional"
)
image = modeling.generate("Sneaker", model)
modeling.show_image(image)
Sample output
Usage
This model can generate only these classes [ 'T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat', 'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot' ] Pass the class name to make the model to generate the image
🪪 License
Released under the MIT License — free for commercial and research use.
If you use this model, please cite the repository and give credit to Sherwin Roger.
Citation
@software{fashion_mnist_diffusion_2025,
author = {Sherwin Roger},
title = {Fashion-MNIST Diffusion (Class-Conditional UNet)},
year = {2025},
url = {https://huggingface.co/Sherwinroger002/fashion_mnist_diffusion_class_conditional}
}