HRS10K

该数据集名为HRS10K，是用于高分辨率显著目标检测（High-Resolution Salient Object Detection，简称HRSOD）的最大数据集，包含了10,500张2K-8K分辨率的高质量标注图像。这些图像是从公共网站收集而来，并经过人工标注以产生高质量的遮罩。数据集涵盖了广泛的对象类型，并包含了具有复杂边界、更具挑战性的显著目标。该数据集规模为10,500张图像，其中8,400张用于训练，2,100张用于测试，其任务专注于高分辨率显著目标检测。

The dataset named HRS10K is the largest publicly available dataset for High-Resolution Salient Object Detection (abbreviated as HRSOD). It encompasses 10,500 high-quality annotated images with resolutions spanning from 2K to 8K. These images are sourced from public web platforms and manually annotated to produce high-quality segmentation masks. The dataset covers a diverse set of object categories and incorporates more challenging salient objects with intricate boundaries. Overall, the dataset comprises 10,500 images in total, with 8,400 allocated for model training and 2,100 for model testing, and its core task is high-resolution salient object detection.

搜集汇总

背景与挑战

背景概述

HRS10K是一个高分辨率显著目标检测数据集，包含10,500张图像，分为8,400张训练图像和2,100张测试图像。该数据集是为支持ACM MM2023论文研究而创建的，可通过百度网盘和Google Drive获取。