Sensor Generalization for Adaptive Sensing in Event-based Object Detection via Joint Distribution Training
This work addresses the challenge of limited data variability and lack of parameter analysis for event camera data, which is crucial for developing robust object detection models for researchers and practitioners working with these novel sensors.
This paper investigates the impact of intrinsic sensor parameters on the performance of object detection models trained on event camera data. The authors use their findings to improve the robustness of downstream models, making them more sensor-agnostic.
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.