Conference Program
Registration at the conference venue opens at February 6th, 11:15.
Download the full conference program.
Invited Speaker
 | Gabriel Brostow is a Professor of Machine Vision at University College London. His group specializes in Human-in-the-Loop computer vision, where the applications span different areas of vision applied for scientific exploration, and vision applied for computer graphics. He received his PhD from Georgia Tech, and has worked at Cambridge University and ETH Zurich, before starting his group at UCL. He is an Associate Editor for PAMI, was co-program chair for BMVC 2017, and is co-program chair for ECCV 2022. |
Self-supervision for 3D Shape and Appearance modeling.
A single glimpse is hardly enough to triangulate the 3D shapes of a scene. But many glimpses taken together, can give enough supervision to accomplish interesting tasks, such as depth from a single photo, volume from a single depth, and appearance of objects and scenes from novel viewing angles. In this talk, I will distill the main lessons we have learned recently, in attempting to a) design networks that understand "a bit" about 3D, and to b) train networks to predict depth, or volumes, or appearance, for several application domains. Some details matter, and the data itself is a key ingredient. There is still more exciting work to be done!
This talk will cover equivariance, consistency losses, and some personal views on diversity in predictions.