WhichWaste: Multiclass Waste Sorting Dataset

The dataset contains images of 24 different categories of waste, according to the criteria of waste collection and management. Specifically, the categories, to which different object categories belong, are: bulky Items, cardboard, paper, styrofoam, aluminum-cans, PET, textiles, metal, non-ferrous metal, electric, rubble/stonegood, batteries, lithium batteries, coffee-capsules, screens/displays, books, CD-DVDs, cork, pots/pans, mixed-materials, dangerous-goods, plastic bottles (not-PET), tires, residual waste (trash). The data collection is based on manual collection (field photos), integration of material from publicly available datasets (Imagenet, Taco, TrashNet, MJU-Waste) and open images web-scraping. Overall the dataset contains 14'270 images and it represents one of the biggest open image dataset on the topic waste. The dataset is balanced.

本数据集按照垃圾收集与管理的相关标准，涵盖24类不同垃圾的图像。具体涵盖的垃圾类别如下：大件垃圾、硬纸板、纸张、发泡聚苯乙烯泡沫（保丽龙）、铝制易拉罐、PET（聚对苯二甲酸乙二醇酯）、纺织物、金属、有色金属、电气类垃圾、碎石/石料、普通电池、锂电池、咖啡胶囊、屏幕/显示器、书籍、CD-DVD光盘、软木材质、锅具、混合材质垃圾、危险废弃物、非PET塑料瓶、轮胎、其他生活垃圾（残余垃圾）。数据采集方式包括实地手动拍摄（现场照片采集）、整合公开数据集（Imagenet、Taco、TrashNet、MJU-Waste）中的素材，以及通过网络爬虫抓取开源图像。本数据集总计包含14270张图像，是当前垃圾主题开源图像数据集中规模最大的数据集之一。该数据集的样本分布均衡。