COD10KD, NC4K-D, CAMO-D
收藏资源简介:
该数据集由华中科技大学的研究团队创建,旨在为现实中的伪装物体检测(RCOD)任务提供基准测试。数据集基于现有的COD10K-v2、NC4K和CAMO数据集,通过手动标注边界框和类别标签,生成了COD10KD、NC4K-D和CAMO-D三个新的数据集。这些数据集包含了伪装物体与其背景高度相似的特征,适用于检测任务的评估。数据集的应用领域主要集中在搜索与救援、军事打击等需要精确定位伪装物体的场景,旨在通过优化检测模型的前景与背景识别能力,提升RCOD任务的性能。
This dataset was developed by a research team from Huazhong University of Science and Technology, with the purpose of providing a benchmark for the Real-world Camouflaged Object Detection (RCOD) task. Based on existing datasets including COD10K-v2, NC4K, and CAMO, three novel datasets named COD10KD, NC4K-D, and CAMO-D were created via manual annotation of bounding boxes and category labels. These datasets feature scenarios where camouflaged objects possess high visual similarity to their background, making them ideal for evaluating detection performance. The application scenarios of these datasets primarily cover search and rescue, military strikes, and other scenarios requiring accurate localization of camouflaged objects, aiming to enhance the performance of RCOD tasks by optimizing the foreground-background recognition capabilities of detection models.




