Haiwen (Haven) Feng (Max Planck Institute for Intelligent Systems): “Structural Imagination: From Inverse Graphics to World Simulation”

Speaker: Haiwen (Haven) Feng (Max Planck Institute for Intelligent Systems)

Title: Structural Imagination: From Inverse Graphics to World Simulation

Video: Click here to view

Abstract: Our visual world is rich, fine-grained yet deeply structured. From a single moment, humans can intuitively imagine countless alternative realities by manipulating underlying factors—lighting, geometry, motion, or intent. How can we build a world simulator that imagines in the same controllable, compositional way?
In this talk, I will revisit the long-standing vision of inverse graphics—the effort to infer and generate the structured world behind images—and discuss how it can be redefined in the era of foundation models. I will show how recent advances enable us to generalize inverse graphics beyond fixed representations, toward scalable world simulation that unifies perception and generation within a learned, controllable framework.

Bio: Haiwen (Haven) Feng is a PhD candidate at the Max Planck Institute for Intelligent Systems, co-advised by Michael J. Black and Angjoo Kanazawa, and an incoming postdoctoral fellow at Berkeley AI Research working with Angjoo Kanazawa and Trevor Darrell. His research focuses on inverse graphics, scalable 3D/4D visual representations, and controllable generation with foundation models.