COCO-Tasks
收藏arXiv2025-09-30 收录
下载链接：
https://coco-tasks.github.io/
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为COCO-Tasks，它对COCO数据集进行了重新标注，加入了考虑偏好意识的可供性标签，专注于面向任务的检测和实例分割。该数据集是唯一一个在可供性中涉及实例级偏好的数据集。它利用现有的COCO遮罩，将COCO数据集扩展到了实例分割版本。该数据集包含了14个任务，每个任务都有3600张训练图像和900张测试图像，其规模适用于面向任务的实例分割和检测。
The dataset named COCO-Tasks is a reannotated version of the COCO dataset, which integrates preference-aware affordance labels and focuses on task-oriented detection and instance segmentation. It is the only dataset that covers instance-level preferences in affordance research. Leveraging the existing COCO masks, it extends the COCO dataset to support instance segmentation tasks. The dataset includes 14 distinct tasks, with each task providing 3,600 training images and 900 test images, a scale that is appropriately sized for task-oriented instance segmentation and detection research.
