Understanding the Visual World Through Object Intrinsics

Explore a comprehensive lecture on understanding the visual world through object intrinsics, presented by Jiajun Wu at #ICBS2024. Delve into the intrinsic structure of our visual world, examining how scenes are composed of objects with specific geometry, texture, and material properties, often exhibiting partial symmetry and repetition. Investigate recent efforts in scene and object understanding through the modeling of these intrinsic properties. Learn about the hybrid neuro-symbolic approach, which combines symbolic programs and neural networks to achieve data-efficient learning, better generalization, and robust structural information capture. Discover how this method effectively extracts complex, low-level features from cluttered and noisy visual data while maintaining the expressiveness of neural networks.