Archive
Notes and Cheat Sheet
Review
A sequel to CS 180 from the Fall, CS 280 solidified my CV fundamentals and dove headfirst into research topics. The first half of the class was essentially a beefed up version of 180. Jitendra Malik came in and gave 2 rather insightful lectures about the invention of homogeneous coordinates. Then we covered SfM and neural/transformers based methods. The coverage of flow matching and diffusion was much more rigorous than 180; Prof. Kanazawa taught it based on the ODE and SDE viewpoint rather than invertible networks (thank goodness). As always, it’s great to hear Efros lecture. This time around he hit us with some timely and somewhat philosophical questions about self supervised learning. In the last unit we covered Dust3r, VGGT, World Models, pose estimation, 4D, and some more research-y topics.
Supposedly Alyosha’s eyes are in poor shape so he compensated by becoming a computer vision researcher.
I found it useful to independently record lectures - more so for Angjoo’s lectures.
CS 280 was a fun class, as in 5/5 in fun and 3.5/5 in difficulty. I actually nearly topped the semester-end exam (yes, in a room of ML PhDs!) and would have done so if I didn’t skip the lecture on Dust3r and VGGT (I somehow got all 8 T/F questions wrong by guessing).