3D Gaussian Splatting from Scratch — PyTorch Only

Learn to build a full 3D Gaussian Splatting (3DGS) pipeline entirely in Python and PyTorch — no CUDA, no external libraries. This course walks you through every step of modern neural rendering: parsing COLMAP outputs, implementing 3D Gaussian primitives, creating a trainable radiance model, coding the optimization loop, and building a real-time differentiable renderer for photorealistic results. Ideal for researchers, graduate students, and developers who want a clean, modular PyTorch implementation and a deep understanding of 3DGS without C++ or CUDA. After 7k training iterations, your metrics will match those reported in the original paper. By the end, you’ll have a complete working implementation ready for extension and research use.

Introduction to Rectified Flow and Stable Diffusion 3.5

This course is an introduction to Rectified Flow and Stable Diffusion 3.5 for people who want to understand how modern diffusion models work in practice. We cover the key background ideas, then go through the main concepts behind Rectified Flow and Stable Diffusion 3.5, including architecture and sampling, and implement the core pieces from scratch.