Renato Martins

SVI360 achieves superior omnidirectional video interpolation by effectively managing distortions that plague traditional methods, setting a new standard for immersive video applications.

Synthesizing new viewpoints during training and encoding temporal dynamics with selective state-space models dramatically improves action detection, even when real-world videos have wildly different camera angles.

Zero-shot action recognition gets a boost: this method handles variations in viewpoint and body orientation, a common stumbling block for real-world deployment.