Sensor Generalization for Adaptive Sensing in Event-based Object Detection via Joint Distribution Training
2026-02-26 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors studied special cameras called bio-inspired event cameras, which quickly detect changes in a scene and work well in tricky lighting. These cameras produce new types of data that aren't well understood yet. The authors analyzed how the camera settings affect an AI model's ability to detect objects using this data. Their work also helps make the AI model work better with different types of event cameras.
event camerasbio-inspired sensorsobject detectionasynchronous datahigh dynamic rangemotion blursensor parametersmodel robustness
Authors
Aheli Saha, René Schuster, Didier Stricker
Abstract
Bio-inspired event cameras have recently attracted significant research due to their asynchronous and low-latency capabilities. These features provide a high dynamic range and significantly reduce motion blur. However, because of the novelty in the nature of their output signals, there is a gap in the variability of available data and a lack of extensive analysis of the parameters characterizing their signals. This paper addresses these issues by providing readers with an in-depth understanding of how intrinsic parameters affect the performance of a model trained on event data, specifically for object detection. We also use our findings to expand the capabilities of the downstream model towards sensor-agnostic robustness.