Level 4 · Modern Architectures & Production CV

The last few years of computer vision have moved past CNNs alone: vision transformers, embedding-based retrieval, self-supervised pretraining, and generative models are now part of the everyday toolkit. This level covers all of them at a real, practical level — running real pretrained ViT and embedding models, building a real similarity-search index, and generating real images — alongside the production concerns (robustness, bias, latency at scale, and choosing the right approach for a real problem) that decide whether a vision system that works in a notebook actually works for real users.

By the end of this level you'll have run modern vision architectures locally, reasoned honestly about their failure modes, and built a real, complete, end-to-end computer vision application as this course's capstone.