Skip to content

14/01/2026 4:33 PM - Lecture

⬅️ [14/01/2026 3:01 PM - PACES](<./14_01_2026 3_01 PM - PACES.md>) | ⬆️ [EECS 504](<./README.md>) | [16/01/2026 11:17 AM - Lecture](<./16_01_2026 11_17 AM - Lecture.md>) ➡️

14/01/2026 4:33 PM

PACES grade is the turn in.
The Perusall assignment is for the video and reading combined.

The true distribution of images is a very small (and relatively easy to define) subset of the total set of all images. Of course Jason talks about a manifold immediately. Computer vision models this manifold where actual images lie.

Talk

Do we put lidars on top of the cars that are not at all hidden as a trust thing?

Can be improve stereo vision algorithms by giving hints with monocular depth algorithms

Do you use the same backbone for both segmentation tasks and detection tasks?

Do you estimate the difficulty of the current driving conditions? Not just weather, but where we are? If so what's the dataset for that?

How were thresholds set for performance? What is "good enough"?
The BEV lane detection seems like a complicated system. Does that add more potential failure modes?


⬅️ [14/01/2026 3:01 PM - PACES](<./14_01_2026 3_01 PM - PACES.md>) | ⬆️ [EECS 504](<./README.md>) | [16/01/2026 11:17 AM - Lecture](<./16_01_2026 11_17 AM - Lecture.md>) ➡️