REEC, a real-world event-based dataset for exposure correction

Existing exposure correction datasets, typically captured using conventional low dynamic range sensors, often suffer from irreversible information loss in severely under- or over-exposed regions. This limitation hinders the accuracy and robustness of image reconstruction, particularly in complex real-world scenarios. To address this challenge, we leverage the high dynamic range sensing capability of event cameras to better capture illumination variations under extreme exposure conditions. We introduce REEC, a large-scale, real-world dataset containing 11,100 spatially and temporally aligned event-image pairs, collected using a precisely calibrated, high-resolution hybrid-camera system. The dataset covers a wide range of exposure levels, scene types, and lighting conditions, serving as a resource for event-driven low-level vision research and applications.