Meeting ID: 862 3823 3298
Title: 3D from one or more Images
Abstract: The world we live in is incredibly diverse, comprising of over 10k natural and man-made object categories. To allow scalable 3D prediction for such generic objects, recent approaches have striven to learn 3D from category-level segmented image collections. However, these methods learn independent category-specific models from scratch, often relying on adversarial or template-based priors to regularize learning. In this talk, I will present a simpler alternative for scalable 3D reconstruction — learning a unified model across 150 categories while using synthetic 3D data on some categories to help regularize learning for others. Moving beyond single-view reconstruction, I will also show how sparse multi-view collections can allow us to infer fine instance-level 3D shapes for generic objects using as few as 8 images.
Bio: Shubham Tulsiani is an Assistant Professor in the Robotics Institute at CMU. Prior to this, he was research scientist at Facebook AI Research (FAIR) and received a PhD. in Computer Science from UC Berkeley in 2018. He is interested in building perception systems that can infer the spatial and physical structure of the world they observe.