COOOLER

该数据集名为COOOLER，它扩展了Challenge-of-Out-of-Label（COOOL）基准，通过为危险标注提供完整的自然语言描述，提高了自动驾驶系统中危险检测的质量和有效性。数据集包含了经人工标注的危险标签和描述，这些描述通过余弦相似度指标进行评估，并进行了水印去除和降噪等增强处理。该数据集的规模包括200个短视频片段，其任务是进行危险检测和标注评估。

Named COOOLER, this dataset extends the Challenge-of-Out-of-Label (COOOL) benchmark by providing complete natural language descriptions for hazard annotations, thereby enhancing the quality and effectiveness of hazard detection in autonomous driving systems. It contains manually annotated hazard labels and descriptions, where the descriptions are evaluated using the cosine similarity metric and processed with enhancement operations including watermark removal and noise reduction. The dataset comprises 200 short video clips, with its core tasks focusing on hazard detection and annotation evaluation.