COCO Irregular Occlusion Dataset(CIOD)

The COCO Irregular Occlusion Dataset (CIOD) is a derived dataset created from the COCO dataset, focusing on irregularly-shaped occluders extracted from object masks. These occlusion samples are intended to simulate real-world complex occlusions and are specifically designed for use in occluded person re-identification (ReID) tasks. CIOD provides diverse and realistic occlusion patterns to enhance model robustness in challenging visual scenarios.

COCO不规则遮挡数据集（COCO Irregular Occlusion Dataset，CIOD）是基于COCO数据集构建的衍生数据集，其聚焦于从目标掩码中提取的不规则形状遮挡物。此类遮挡样本旨在模拟真实世界中的复杂遮挡场景，专为遮挡场景下的行人重识别（ReID）任务设计。CIOD提供了多样化且极具真实感的遮挡模式，用以提升模型在高难度视觉场景中的鲁棒性。