Inverse Graphic

Inverse graphics aims to reconstruct a 3D scene's physical properties (geometry, materials, lighting) from a 2D image, effectively reversing the image formation process. Current research focuses on improving robustness and efficiency through probabilistic inference methods, leveraging large language models for scene understanding, and employing novel differentiable rendering techniques like shadow mapping and neural radiance fields (NeRFs). These advancements enable applications such as realistic image relighting, free-viewpoint rendering, and robust 6D pose estimation, pushing the boundaries of computer vision and graphics.