Frag125 PKU Dataset

Wen Dong, Haiyang Mei, Yinglian Ji, Yutong Jiang, Ziqi Wei, Shengfeng He, Xin Yang

You Only Look Intensity Once: Event-Driven Long-Term High-Speed Object Detection

International Journal of Computer Vision 134.4 (2026): 149.

https://link.springer.com/article/10.1007/s11263-026-02749-8

Introduction

We use the PKU DAVIS SOD dataset (Li et al., 2023), which provides synchronized event streams and RGB frames under normal illumination, low-light, and motion blur conditions. To address missing event segments, incomplete annotations, and ambiguous labels, we construct a high-quality subset (termed as Frag125 PKU) by splitting each sequence into 125-frames (5-second) fragments and excluding any segments with annotation inconsistencies or

event-data gaps, while maintaining the original train/val/test splits.