Research highlights:
Multisensory supervision: Today's computer vision methods need human supervision, such as object labels, to learn about the world. Humans, on the other hand, learn a great deal from associations between senses: vision trains hearing, touch trains vision, etc. Inspired by this idea, I've been developing models that learn about the world by finding structure in multimodal sensation—especially "self-supervised" computer vision methods that learn from sound.
Touch sensing: When humans interact with objects, they use many modalities: touch sensing helps us position our hands and select which forces to exert, while vision is useful for choosing where to grip. Recently, I've been developing multimodal methods for robotic grasping.
Spotting fake images: Computer vision researchers face a dilemma: as our methods get better, so do the tools for malicious image manipulation. To address this growing issue, I've been developing methods for detecting fake images.
3D reconstruction: To interact with the world, we need to know not just what is in a scene, but what is where. I've developed methods for reconstructing scenes in 3D from multiple visual cues.