MCOD
收藏资源简介:
MCOD数据集是首个专门为多光谱伪装目标检测设计的基准数据集。该数据集包含了8种具有挑战性的属性,如小目标尺寸、极端光照条件等,以更准确地反映现实世界的困难。MCOD数据集覆盖了从城市公园到草地等广泛的自然环境,增强了数据集的泛化性和实用性。所有图像都进行了像素级精确标注,并通过多轮人工审核确保了高质量。MCOD数据集的引入旨在促进从多光谱数据中学习更具辨别力的伪装特征,推动COD的理论理解和实际应用。
The MCOD dataset is the first benchmark dataset specifically designed for multispectral camouflaged object detection (COD). It includes eight challenging attributes such as small target size and extreme lighting conditions to more accurately reflect real-world difficulties. The MCOD dataset covers a wide range of natural environments ranging from urban parks to grasslands, enhancing its generalization and practicality. All images are annotated with pixel-level precision and go through multi-round manual reviews to ensure high quality. The introduction of the MCOD dataset aims to promote the learning of more discriminative camouflaged features from multispectral data, and advance the theoretical understanding and practical applications of COD.




