PEOD: A Pixel-Aligned Event-RGB Benchmark for Object Detection under Challenging Conditions

The introduction of the PEOD dataset marks a pivotal moment in the field of object detection, particularly under challenging conditions. With a resolution of 1280 x 720, it surpasses previous datasets that were limited to lower resolutions, thus enabling more accurate evaluations of detection algorithms. PEOD comprises over 130 spatiotemporal-aligned sequences and 340,000 manual bounding boxes, with a notable 57% of the data collected in adverse conditions like low light and high-speed motion. This dataset allows researchers to benchmark 14 different detection methods across various input configurations, revealing that while fusion-based models perform excellently in standard conditions, the top event-based model excels in low-illumination scenarios. This highlights the limitations of current fusion techniques when faced with severely degraded frame modalities. The PEOD dataset not only enhances the capabilities of object detection systems but also sets a new standard for future resear…