Diffusion Models

This repository contains the code and results for our term project in the UMC203: Introduction to AI and ML course at IISc.

📌 Contributors

• Om Prakash Choudhary
• Himesh Kumar Anant
• Jithendra Rao Kasibhatla
• Keval Pithadiya

---

🧪 Results

For a detailed explanation of the methodology, training setup, and results, check out the full report:
📄 Read the Report (PDF)

CelebA 128x128 – Noise Prediction

DDPM – Noise Prediction

DDIM – Noise Prediction

---

CIFAR-10 32x32

DDPM – Noise Prediction	DDIM – Noise Prediction
DDPM – Clean Image Prediction	Score Matching

---

Miscellaneous

Sample Quality vs Sampling Steps

Latent Interpolation (CelebA 128x128)

---

🧠 Models Trained

We trained Denoising Diffusion Probabilistic Models (DDPMs) using Hugging Face's diffusers library for both noise prediction and clean image prediction objectives.

📊 Datasets and Resolutions

• CIFAR-10: 32×32
• CelebA: 32×32 and 128×128
• MNIST: 32×32

🔗 Pretrained Models

• 📁 Google Drive Folder
• 🤗 HuggingFace Model Hub

---

🛠️ Setup Instructions

Requirements

HuggingFace diffusers Library

pip install diffusers[training]

Python Version Used: 3.13.2