SK-LARGE

该数据集，即 SK-LARGE，用于评估自然图像中的骨骼提取方法，其中包含从 MS COCO [1] 中选择的 1491 张图像，746 幅用于训练，745 幅用于测试。

SK-LARGE中的每个图像都有一个对应的骨架图，使用[2]中介绍的方法进行提取。每个 groundtruth 由一个 .mat 文件表示，其中有两个字段：

edge：物体的轮廓（边缘）图，1：轮廓； 0：非轮廓；
对称性：物体的骨架（对称）图，1：骨架； 0：非骨架。

This dataset, namely SK-LARGE, is designed to evaluate skeleton extraction methods for natural images. It comprises 1491 images selected from MS COCO [1], with 746 allocated for training and 745 for testing.

Each image in SK-LARGE has a corresponding skeleton map extracted using the method introduced in [2]. Each ground truth is represented by a .mat file containing two fields:
- edge: An object contour (edge) map, where 1 denotes contour pixels and 0 denotes non-contour pixels;
- symmetry: An object skeleton (symmetry) map, where 1 denotes skeleton pixels and 0 denotes non-skeleton pixels.

搜集汇总

背景与挑战

背景概述

SK-LARGE是一个公开的图像CV物体骨架检测数据集，由OpenDataLab发布，具体数据量和应用场景未明确说明。