COCO-Occ
收藏资源简介:
COCO-Occ数据集是由华威大学计算机科学系的研究团队基于COCO数据集创建的,旨在解决全景分割和图像理解中的遮挡问题。该数据集包含35,000张图像,分为30,000张训练图像和5,000张测试图像,每张图像都经过手动标注,分为低、中、高三个遮挡等级。数据集的创建过程包括使用COCO API叠加多边形掩码,并根据遮挡率手动分类图像。COCO-Occ数据集主要应用于全景分割任务,旨在提高模型在不同遮挡程度下的表现,特别是在高遮挡情况下的性能。
The COCO-Occ dataset was developed by the research team from the Department of Computer Science at the University of Warwick based on the COCO dataset, aiming to address occlusion issues in panoptic segmentation and image understanding. This dataset contains 35,000 images, divided into 30,000 training images and 5,000 test images. Each image has undergone manual annotation with three occlusion levels: low, medium, and high. The dataset construction process uses the COCO API to overlay polygonal masks, and manually classifies images based on their occlusion rates. The COCO-Occ dataset is primarily applied to panoptic segmentation tasks, with the goal of improving model performance under varying degrees of occlusion, especially in high-occlusion scenarios.




