ddpm-unconditional-faces-64x64

<h2 align="center" style="font-size: 2.8em; font-weight: 800; margin-bottom: 20px;"> Unconditional DDPM for Face Generation </h1>

---

<table align="center" border="0" style="border-collapse: collapse; border: none; width: 100%;"> <tr align="center" valign="middle" style="border: none;"> <td width="33%" style="border: none;"> <img src="https://uxwing.com/wp-content/themes/uxwing/download/brands-and-social-media/pytorch-icon.svg" width="100" height="100" style="object-fit: contain;" alt="PyTorch Framework"/> <br><br> <b>Built with PyTorch</b> </td> <td width="33%" style="border: none;"> <a href="https://colab.research.google.com/drive/1DEuyTrVwfmWNaGmxyhd3ink-_3uF8a2A?usp=sharing" target="_blank"> <img src="https://colab.research.google.com/img/colab_favicon_256px.png" width="100" height="100" style="object-fit: contain;" alt="Google Colab"/> <br><br> <img src="https://img.shields.io/badge/Open%20Notebook-Google%20Colab-F9AB00?style=for-the-badge&logo=googlecolab&logoColor=white" alt="Open Notebook"/> </a> </td> <td width="33%" style="border: none;"> <img src="https://huggingface.co/front/assets/huggingface_logo-noborder.svg" width="100" height="100" style="object-fit: contain;" alt="Hugging Face"/> <br><br> <b>Hosted on Hugging Face</b> </td> </tr> </table>

---

Sample Visualizations

Model Generations (Sampling)

Reverse Diffusion Trajectory

---

Code to Run Inference

import matplotlib.pyplot as plt
from huggingface_hub import hf_hub_download
import torch

repo_id = "SavirD/ddpm-unconditional-faces-64x64"

hf_hub_download(repo_id=repo_id, filename="model.py", local_dir=".")
hf_hub_download(repo_id=repo_id, filename="pipeline.py", local_dir=".")
weights_path = hf_hub_download(repo_id=repo_id, filename="model_epoch_50.pth")

from pipeline import DDPMFacePipeline

pipeline = DDPMFacePipeline(checkpoint_path=weights_path)
images = pipeline.generate(num_samples=4)

fig, axes = plt.subplots(1, 4, figsize=(12, 3))
for idx, ax in enumerate(axes):
    ax.imshow(images[idx])
    ax.axis('off')
plt.suptitle("Faces via Unconditional DDPM", fontsize=14)
plt.show()