Diffusion Models Beat GANs on Image Synthesis | ML Coding Series | Part 2

❤️ Become The AI Epiphany Patreon ❤️ 👨‍👩‍👧‍👦 Join our Discord community 👨‍👩‍👧‍👦 4th video in the ML coding series! In this one I continue explaining diffusion models! I cover the “Diffusion Models Beat GANs on Image Synthesis“ paper and the code behind it. I focus on how classifier guidance works. I cover both the training of the noise-aware classifier as well as the actual sampling (the mean shift method). I also walk you through a minor bug in their code. Let me know how you find this format! ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ ✅ GitHub: ✅ My issue: ✅ Paper: ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ ⌚️ Timetable: 00:00:00 Intro 00:01:30 Paper overview part - U-Net architecture improvements 00:05:38 Classifier guidance explained 00:15:18 Intuition behind classifier guidance 00:20:10 Scaling classifier guidance 00:24:10 Diversity vs quality tradeoff and future work 00:26:15 Coding part - training a noise-aware classifier 00:35:35 Main training loop 00:44:26 Visualizing timestep conditioning 00:46:00 Sampling using classifier guidance 00:52:35 Core of the sampling logic 00:59:20 Shifting the mean - classifier guidance 01:05:03 Minor bug in their code and my GitHub issue 01:07:53 Outro ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ 💰 BECOME A PATREON OF THE AI EPIPHANY ❤️ If these videos, GitHub projects, and blogs help you, consider helping me out by supporting me on Patreon! The AI Epiphany - One-time donation - Huge thank you to these AI Epiphany patreons: Eli Mahler Petar Veličković ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ 💼 LinkedIn - 🐦 Twitter - 👨‍👩‍👧‍👦 Discord - 📺 YouTube - 📚 Medium - 💻 GitHub - 📢 AI Newsletter - ▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬ #diffusion #generativemodeling #coding