RClicks
收藏资源简介:
RClicks数据集是由AIRI和莫斯科国立大学合作创建的,旨在模拟真实的用户点击行为,以评估交互式分割方法的性能。该数据集包含475,544条用户点击数据,涵盖了GrabCut、Berkeley、DAVIS、COCO-MVal和TETRIS等多个图像分割数据集。数据集的创建过程包括通过众包平台Toloka AI收集用户点击,并使用多种显示模式来减少展示偏差。RClicks数据集主要应用于交互式分割领域,旨在解决现有方法在实际应用中性能下降的问题,特别是通过模拟真实用户点击来评估方法的鲁棒性。
The RClicks dataset was collaboratively created by AIRI and Lomonosov Moscow State University, aiming to simulate real user click behaviors for evaluating the performance of interactive segmentation methods. It contains 475,544 user click records, covering multiple image segmentation datasets including GrabCut, Berkeley, DAVIS, COCO-MVal, and TETRIS. The dataset was developed by collecting user clicks via the crowdsourcing platform Toloka AI, and multiple display modes were used to reduce presentation bias. Primarily applied in the field of interactive segmentation, the RClicks dataset is designed to address the performance degradation issue of existing methods in real-world scenarios, particularly to evaluate the robustness of methods by simulating real user clicks.




