Intro to AI Image Generation: Diffusion, Math, Code скачать в хорошем качестве

Intro to AI Image Generation: Diffusion, Math, Code

My Core Diffusion repo: https://github.com/InexperiencedMe/co... This time, my journey to Master Machine Learning took me to learn Diffusion Models. From scratch. And I explain everything along the way. We start with investigating image generation as a problem, then we talk about GANs that dominated image generation for some years, then autoregression, then diffusion through the DDPM paper. I’m even surprised myself how long it took to understand the paper because as it usually is, we have no idea how demanding the journey is until we take it. And I had to understand EVERYTHING there, so I read the paper, I derived all crucial equations, then implemented them in practice. What a trip, what a video. In the video I mention this video by Depth First explaining Langevin Dynamics in the context of Diffusion Models:    • More Than Image Generators: A Science of P...   Also, when I say about great derivations video, one of the best videos is:    • Diffusion Models | Paper Explanation | Mat...   But to be honest, I watched like every video on YouTube about the DDPM paper and checked a lot of articles, so there were many more of them, that I synthesized into this one long video. Chapters: 00:00 Hook 01:46 Exploring the problem 06:23 GANs 09:48 Autoregression 20:23 Diffusion 23:44 DDPM: Abstract 30:12 DDPM: Introduction 35:42 First graphs and notation 52:11 DDPM: First equations 59:00 Variational Bound Derivation 1:15:06 Forward Skip Derivation 1:21:17 KL Loss Derivation 1:43:35 True Reverse Diffusion Derivation 1:48:10 L_T 1:49:50 Main Loss Derivation 2:06:18 L_0 2:09:19 Simplified Loss 2:12:10 Training and Sampling algorithms 2:18:06 Core Diffusion code