ECE Department Seminar: Anand Bhattad

Description

Anand Bhattad, an assistant professor of computer science and a member of the Data Science and AI Institute at the Johns Hopkins University, will give a talk titled "The Three P's of Modern Computer Vision: Beyond Recognition, Reconstruction, and Reorganization" for the Electrical and Computer Engineering Department.

Abstract:

For decades, computer vision has been defined by three grand challenges, elegantly summarized by Jitendra Malik et al. as the "Three R's": Recognition (what is it?), Reconstruction (what is its 3D shape?), and Reorganization (what belongs together?). While this framework has guided the creation of powerful computer vision systems, the next frontier lies in moving from static description to dynamic, interactive understanding. This talk presents our research on a new paradigm, "Pixels -> Perception -> Physics," that leverages generative visual models to learn how to see and act in the world.

First, I will demonstrate how we can create interactive 3D worlds from single images and learn their appearance as latent variables without supervision. Next, I will show how we push into intuitive physics by exploiting counterfactual reasoning with generative models to play Visual Jenga—a new scene understanding task that we propose. I will conclude by examining the key failures of these generative models on physical laws like projective geometry and gravity.