Visible-Thermal Cross-Modality Class-Incremental Learning

We construct a cross-modality CIL benchmark based on the requirements of numerous real-world scenarios. The datasets in this benchmark consider the diversity of scene complexity and environmental factors, covering urban roads, residential areas, and other scenarios, spanning seasons, weather conditions, daylight, and nighttime. We conduct extensive experiments on two cross-modality CIL benchmark datasets, i.e., LasHeR-CIL and VTDV-CIL.