René Schuster

Overcome the Achilles' heel of event-based vision—its sensitivity to sensor parameters—with a training scheme that achieves sensor-agnostic object detection.

Fusing monocular RGB images with sparse LiDAR data lets you estimate scene flow more accurately and efficiently than using either modality alone.

Jointly calibrating LiDAR, RGB, and event data is now possible with LiREC-Net, a single network that rivals the accuracy of specialized bi-modal methods.

Forget retraining? This new method uses SAM to segment anything and then incrementally learns semantics, beating standard continual learning approaches *without any training*.