Distinctions-646

该数据集名为“Distinctions-646”，包含了训练集和测试集两部分。训练集中包含596个前景物体及其对应的真实alpha蒙版，而测试集则包含50张前景图像及其相应的alpha蒙版。在训练集中，每张前景图像都与100张背景图像进行了组合，而在测试集中，每张前景图像则与20张背景图像进行了组合。该数据集的任务是图像抠图。

The dataset is named "Distinctions-646", which comprises two subsets: the training set and the test set. The training set contains 596 foreground objects along with their corresponding ground-truth alpha mattes, while the test set includes 50 foreground images and their respective alpha mattes. For the training set, each foreground image is combined with 100 background images, whereas for the test set, each foreground image is paired with 20 background images. The task of this dataset is image matting.

搜集汇总

背景与挑战

背景概述

Distinctions-646是一个专为无trimap图像抠图研究设计的数据集，包含646张带有精细手动标注alpha蒙版的前景图像。该数据集旨在支持学术研究，需通过邮件联系作者获取，且仅限非商业用途。