Waste datasets review

List of datasets with any kind of litter, garbage, waste and trash. Created during the detectwaste.ml project

Today, more than 300 million tons of plastic are produced annually. Plastic is everywhere and we constantly use it in our daily life.

The idea of detect waste project is to use Artificial Intelligence to detect plastic waste in the environment. Our solution will be applicable for video and photography. Our goal is to use AI for Good.

Visit majsylw/litter-detection-review to see broader review of papers, projects and other resources concering the problem of litter in an environment.

Contributing

Feel free to add issue with short description of new dataset or create a pull request - add the new dataset to the table or fill missing description.

Summary

Name	No. categories	No. subcategories	No. images	Annotation	Comment	Website	Description
TrashCan 1.0	3	34	7 212	Instance-Segmentation	Underwater images	website	✔️
Trash-ICRA19	3	34	5 700	Detection	Underwater images	website	✔️
TACO	28	60	1 500	Segmentation	Waste in the wild	website	✔️
TACO bboxes	7	60	WIP	Detection	Waste in the wild	WIP	✔️
UAVVaste	1	-	772	Segmentation	Drone dataset	github	✔️
Trashnet	6	-	2 527	Classification	Clear background	github	✔️
WaDaBa	8	color,size, shape, or material	4 000	Classification	Plastic dataset, clear background	website	✔️
GLASSENSE-VISION	7	136	2 000	Classification	Home-supplies, clear background	website	✔️
Waste Classification data	2	-	~25 000	Classification	Scraped from google search	kaggle	✔️
Waste Classification Data v2	3	-	~27 500	Classification	Scraped from google search	kaggle	✔️
Waste Images from Sushi Restaurant	16	-	500	Classification	Clear background	kaggle	✔️
Open litter map	11	187	> 100k	Multilabel classification	Waste in the wild	website	✔️
Litter	24	size, shape, or material	~14 000	Detection	Waste in the wild, paid license	website	✔️
Drinking Waste Classification	4	-	9640	Detection	Clear background, (cans and bottles)	kaggle	✔️
waste_pictures	34	-	~24 000	Classification	Scraped from google search	kaggle	✔️
spotgarbage	3	-	~2 400	Classification	Scraped from Bing search	kaggle github	✔️
DeepSeaWaste	5	-	3 055	Classification	Underwater images	kaggle	✔️
MJU-Waste v1.0	1	-	2475	Segmentation	Plain background, indoor RGBD images	github	✔️
Domestic Trash Dataset	10	-	> 9000	Classification/Detection	Waste inn the wild, paid license, 250 images for free	github	✔️
Cigarette butt dataset	1	-	2200	Detection	Waste inn the wild, synthetic images	website	✔️
TrashBox	7	25	17785	Classification/Detection	Scraped from web	github	✔️

Description

TrashCan 1.0

An Instance-Segmentation Labeled Dataset of Trash Observations

7212 images under 3 main categories: bio, trash, unknown. Categories:

bio = turtle, squid, lobster, unknown, jellyfish, stingray, shrimp, crawfish, octopus, shark, shell, crab, starfish, eel
trash = clothing, pipe, bottle, bag, snack_wrapper, glove, tire, can, cup,container, branch, wreakage, tarp, box, hose, rope, hay, net, paper, bucket, wire
unknown Download: Directly from website https://conservancy.umn.edu/handle/11299/214865

Trash-ICRA19:

A Bounding Box Labeled Dataset of Underwater Tras 5,700 underwater images extracted from video https://jungseokhong.github.io/

Download: Directly from website https://conservancy.umn.edu/handle/11299/214366

TACO

Open dataset with 1500 images from 28 categories and 60 detailed sub-categories of waste in the wild. Annotations available in COCO-json.

Download: Directly from website http://tacodataset.org/

TACO bboxes

Additional hand-labelled annotations for images from TACO dataset. There are seven recognized waste categories:

bio: food waste such as fruit, vegetables, herbs, used paper towels and tissues,
glass: glass objects such as glass bottles, jars, cosmetics packaging,
metals and plastic: scrap metal and non-ferrous metal, beverage cans, plastic beverage bottles, plastic shards, plastic food packaging, or plastic straws,
non-recyclable: residual rubbish such as disposable diapers, pieces of string, polystyrene packaging, polystyrene elements, blankets, clothing, or used paper cups,
other: construction and demolition, large-size waste (e.g. tires), used electronics and household appliances, batteries, paint and varnish cans, or expired medicines,
paper: paper, cardboard packaging, receipts, newspapers, catalogues, and books,
unknown waste: (highly decomposed and hard-to-recognize litter),
and extra class background label without any litter: a sidewalk, a forest path, a lawn

Read more about it in the paper Deep learning-based waste detection in natural and urban environments,.

Download: Directly from detect waste repository

UAVVaste

Drone rubbish detection intelligent technology The UAVVaste dataset consists to date of 772 images and 3716 annotations. The main motivation for creation of the dataset was the lack of domain-specific data. The datasets that are widely used for object detection evaluation benchmarking. The dataset is made publicly available and is intended to be expanded.

Avaiable annotations for Detection and Segmentation https://github.com/UAVVaste/UAVVaste

Download: Directly from annotations json on github https://github.com/UAVVaste/UAVVaste

Trashnet

The dataset spans six classes: glass, paper, cardboard, plastic, metal, and trash. Currently, the dataset consists of 2527 images:

501 glass
594 paper
403 cardboard
482 plastic
410 metal
137 trash

Download: Directly from github https://github.com/garythung/trashnet

also is known as Garbage Classification Data

The Garbage Classification Dataset contains 2467 images from 6 categories: cardboard (393), glass (491), metal (400), paper(584), plastic (472) and trash(127).

Download: Directly from kaggle https://www.kaggle.com/asdasdasasdas/garbage-classification

Plastic Waste DataBase of Images – WaDaBa

4000 images with detailed description of a plastic type (PET, PP, PE-HD...), object color, deformation level, dirtiness and others. [classification]

The object were put on the research position and next photographed with first and second type of light. There were series carried out of 10 photographs with differ in the angle of the turnover for every object (in the vertical axis). Next the object was damaged to varying degrees: small, medium and large. For each type of destruction have been made 10 photographs. So considering all variants for every object 40 photographs were taken, multiplying it by the number of objects, 4 000 of photographs were created in the database.

Download: Images free-to-download directly from website. Annotations available after signing license http://wadaba.pcz.pl/#download

GLASSENSE-VISION

Home-supplies classification. It is not strict litter dataset but it gathers over 2000 images with objects well-spareted from background. Covers 7 main categories of (Banknotes, Cereals, Medicines, Cans, Tomato sauces, Water bottle, Deodorant stick) and 136 subcategories.

Glassense-Vision is a set of data we acquired and annotated to the purpose of providing a quantitative and repeatable assessment of the proposed method. The dataset includes 7 different use cases, meaning different object categories, where for each one of them we provide training (reference images used also to build dictionaries) and test images. All images in the dataset are manually annotated. The different use cases (object categories) can be grouped in three main geometrical types:

Download: http://www.slipguru.unige.it/Data/glassense_vision/

Waste Classification data

Over 25k images already divided into training data - 22564 images and test data - 2513 images. Two main categories: Organic and recyclable

Download: Directly from kaggle https://www.kaggle.com/techsash/waste-classification-data

Waste Classification Data v2

A variation about the Waste Classification data: extended by the new category "N" - Nonrecyclable added.

Over 25k images already divided into training data - 22564 + 2508 (N) images and test data - 2513 images + new 397 from category nonrecyclable. Three main categories: Organic (O) and recyclable (R), and nonrecyclable (N). TRAIN folder contains 2508 images in the "N" directory. The TEST folder contains 397 images in the "N" directory.

Download: Directly from kaggle https://www.kaggle.com/sapal6/waste-classification-data-v2

Open litter map

The biggest dataset with over 100k images in total with 11 main categories and 187 subcategories.[multilabel] [classification] https://openlittermap.com/

Download: Only from json with scraper - detectwaste scraper

Litter

The Litter dataset contains 14k images with 20k annotations (bounding boxes) and 24 classes. Each class represents an object (cup), while subclasses determine its size, shape, or material (long paper cup/short paper cup).

Download: After buying a license https://www.imageannotation.ai/litter-dataset

Drinking Waste Classification

The dataset contains ~10k images grupped by 4 classes of drinking waste: Aluminium Cans, Glass bottles, PET (plastic) bottles and HDPE (plastic) Milk bottles. Pictures were taken with 12 MP phone camera as a part of final year Individual Project at University College London. The dataset used parts of manually collected images from TrashNet.

Download: Directly from kaggle https://www.kaggle.com/arkadiyhacks/drinking-waste-classification

waste_pictures

The dataset contains ~24k images grupped by 34 classes of waste for classification purposes. The images were divided into train and test subsets.

Download: Directly from kaggle https://www.kaggle.com/wangziang/waste-pictures

spotgarbage - GINI dataset

The Garbage in Images (GINI) dataset with 2561 images with unspecified resolution, 1496 images were annotated by bounding boxes (one class - trash). Bing Image Search API was used to create their dataset.

Download: Directly from github https://github.com/spotgarbage/spotgarbage-GINI

DeepSeaWaste

This dataset consists of ~3k images divided by 4 categories, and taken under water. In csv file annotations were provided as:

source url of picture,
waste category,
date of taking the picture,
the place and depth at which the waste was found,
information whether it contains living organisms and sediments stuff,
information if this is some plastic bag.

Download: Directly from kaggle https://www.kaggle.com/henryhaefliger/deepseawaste

MJU-Waste v1.0

This dataset was created by capture collected waste items from a university campus in a lab background (people hold waste items in their hands). All images in the dataset are captured using a Microsoft Kinect RGBD camera. All annotations are provided in PASCAL VOC and COCO format.

MJU-Waste v1, contains 2475 co-registered RGB and depth image pairs. Images are randomly splited into a training set, a validation set and a test set of 1485, 248 and 742 images, respectively. Authors used single class label for all waste objects.

Download: From Google Drive link placed on https://github.com/realwecan/mju-waste/

Domestic Trash Dataset

Domestic Trash Dataset consists of images of domestic common trash objects. Images were captured and crowdsourced under wide variety of lighting conditions, weather, indoor and outdoor. This dataset can be used for make trash/litter detection models, eco-friendly alternative suggestions, carbon footprint generation etc.

Dataset Features

Various trash object classes
Has material labels
Captured by 5000+ unique users
Highly diverse and HD
Various lighting conditions
Indoor and Outdoor scenes

Dataset Format

Classification and detection annotations available
COCO, PASCAL VOC and YOLO formats
Approx. 9000+ unique images and growing
Only 250 images for free avaiable on kaggle

Download Images available for download after buying a license. Contact them from their support details at: https://github.com/datacluster-labs/Datacluster-Datasets

Cigarette butt dataset

This dataset consists of a set of 2200 synthetically composed images of cigarettes on the ground. It is designed for training CNNs (convolutional neural networks). You must read and accept the terms of the Non-Commercial, Educational License Agreement to download and use its content.