Josh McDermott continues his lecture on mid-level vision for MIT's course 9.35, Perception, spring 2024. He examines how the brain infers the contours and colors of objects from incomplete or ambiguous visual input, building on concepts introduced in the previous session. The lecture works through the computational problems the visual system must solve to segment scenes into distinct objects and assign stable surface properties to them despite changes in lighting and occlusion. Drawing on psychophysics and perceptual demonstrations, McDermott connects mid-level grouping processes to how humans experience coherent, stable objects rather than raw patches of light. Running 73 minutes, this is one lecture in MIT OpenCourseWare's full recording of the Perception course, aimed at students already familiar with earlier material on low-level vision and contour and color inference.