AIM-500

该数据集名为AIM-500，是首个包含500张多样化高分辨率自然图像的图像抠图测试集，每张图像都附有人工标注的alpha遮罩，涵盖了多种类型，包括人像、动物、透明物体、植物、家具、玩具和水果等。这些图像均从免费授权的网站收集而来，每张图像的短边至少为1080像素。此外，所有的人像图像都不包含可识别信息。该数据集的规模为500张图像，其任务是图像抠图。

Named AIM-500, this dataset is the first image matting test set consisting of 500 diverse high-resolution natural images. Each image is accompanied by manually annotated alpha masks, spanning multiple categories such as portraits, animals, transparent objects, plants, furniture, toys, fruits, and others. All images are sourced from royalty-free websites, with the shorter side of each image being at least 1080 pixels. Moreover, all portrait images included in the dataset do not contain any identifiable information. This dataset comprises 500 images in total, and its target task is image matting.

搜集汇总

背景与挑战

背景概述

AIM-500是一个包含500张高分辨率自然图像的数据集，涵盖多种类型和类别，用于评估自动图像抠图模型的泛化能力。数据集提供了手动标注的alpha遮罩，支持对抠图技术的深入研究。