Biological Computer Vision: an AI Odyssey (Math.AI 4 CV) is a graduate course at IMPA, running from March to June 2027, taught by Luiz Velho and Ricardo Motta with Rafael Romeiro as assistant. The course looks at 3D perception through the lens of biological vision, asking how an active, foveated observer “sees by looking.” It starts from the fundamentals of multiview computer vision and binocular stereo reconstruction, then turns to the physiology of the human eye, foveated vision, and the role of eye movements and vergence in stereoscopic perception. From there it builds toward computational models: a bio-inspired Active Bayesian stereo architecture (the BioEye experiment), Active Foveal 3D Vision (F3D), and a comparative analysis that leads to a unified architecture for biological 3D vision.
The course is also a case study in doing research with AI as a collaborator. Alongside the technical content, it follows the project’s history, key decisions, and open directions, and it shows how AI agents contributed to the design, implementation, and analysis. Students work with synthetic environments built in Blender, including foveated variable-resolution rendering and 360° scene assets, which serve as a controlled testbed for the vision models. Each lecture comes with technical notes, slides, and, where relevant, open-source GitHub repositories and video demos, so students can reproduce and extend the experiments.
(by Claude Chat, Anthropic)