Shape from Shading: Recovering 3D Information from 2D Images

This lecture discusses the process of recovering three-dimensional shape information from two-dimensional images, focusing on the concept of shape from shading. The instructor begins by explaining the transition from 2D images to 3D shapes, emphasizing the importance of contours, textures, and segmentation in building objects. The lecture highlights how shading can provide insights into surface normals, which are crucial for understanding the shape of objects. Various shading models, including the Lambertian model, are introduced to explain how light interacts with surfaces. The instructor also addresses the challenges of shape recovery, such as the need for boundary conditions and the complexities introduced by shadows and specularities. The lecture concludes with a discussion on modern techniques, including deep learning approaches that enhance traditional methods, allowing for better shape recovery even in the presence of ambiguities. The importance of understanding the physics behind these techniques is emphasized, showcasing their application in real-world scenarios such as medical imaging and photometric stereo.