Machine Learning for Inverse Graphics
MIT OpenCourseWare offers this advanced course sitting between computer vision, computer graphics, and geometric deep learning. The subject is inverse graphics: given an image, how does a machine work backward to recover the 3D scene that produced it? Lectures build up from how cameras project a 3D world onto a 2D sensor, then move into representations that let neural networks store and manipulate 3D scenes, including techniques for reconstructing full scenes from a single photograph. Later material covers self-supervised training methods that avoid the need for large labeled 3D datasets, and the theoretical question of what guarantees a model can generalize beyond the scenes it was trained on. Materials follow MIT's standard OpenCourseWare format of lecture notes and readings tied to the Electrical Engineering and Computer Science department, released under a Creative Commons license for self-study. No tuition or certificate fee applies since OpenCourseWare provides no credential, only the course content itself for independent learners already comfortable with linear algebra and machine learning fundamentals.