Better and Faster: Adaptive Event Conversion for Event-Based Object Detection

Event cameras are a kind of bio-inspired imaging sensor, which asynchronously collect sparse event streams with many advantages. In this paper, we focus on building better and faster event-based object detectors. To end, first propose computationally efficient representation Hyper Histogram, adequately preserves both the polarity temporal information events. Then devise an Adaptive Conversion module, converts events into Histograms according to density via adaptive queue. Moreover, introduce novel augmentation method Shadow Mosaic, significantly improves sample diversity enhances generalization ability detection models. We equip our proposed modules three representative models: YOLOv5, Deformable-DETR, RetinaNet. Experimental results datasets (1Mpx, Gen1, MVSEC-NIGHTL21) demonstrate that approach outperforms other state-of-the-art methods by large margin, while achieving much running speed (< 14 ms < 4 for 50 data 1Mpx Gen1 datasets).