More than 50 object detection data sets from different industries

Time:2021-4-29

By Abhishek annamraju
Compile Flin
Source: Media

Computer vision is a rapidly developing field. Every day, a large number of new technologies and algorithms appear in different conferences and journals. When it comes to target detection, you will learn a lot of algorithms in theory, such as fast RCNN, mask RCNN, Yolo, SSD, retinnet, cascade RCNN, peleenet, efficient det, cornernet. This list of algorithms can never be finished!

It’s always helpful to consolidate your learning experience by applying it to different datasets!!!

In this way, you tend to have a better understanding of the algorithms and an intuitive understanding of which algorithms can run on which data sets.

We have compiled a list of object detection, image segmentation and action recognition datasets in the open source team of monk computer vision org, and created a short tutorial for each object, so that you can use these datasets and try different object detection algorithms

The following is a brief list of object detection datasets, brief details about them, and steps to use them. Data sets come from the following areas:

Agriculture
Advanced driver assistance and automatic driving vehicle system
Fashion, retail and marketing
Wildlife
Sports
Satellite imaging
Medical imaging
Safety and monitoring
Underwater imaging

… and more!!!!!

A complete list of instructions and training codes can be found on GitHub

Data sets related to agriculture

A) Winegraph detection data set

*Objective: to detect grape clusters in vineyards

*Application: monitor growth and analyze yield

*Details: 300 images with 4400 bounding boxes for 5 grape categories

*How to use data set and yolov3 pipeline to build custom detector

B) Global wheat detection data set

*Objective: to detect wheat crops in the field

*Application: monitor growth and analyze yield

*Details: 3430 image with 100k + annotation

*How to build a custom detector using data set and efficient det-d4 pipeline

Advanced data set for driver assistance and autopilot vehicle systems

A) Lisa traffic sign detection data set

*Objective: to detect and classify traffic signs in dash cam images

*Application: traffic sign recognition is the rule setting procedure of automatic driving

*Details: there are 7855 comments on 6610 frames of 47 U.S. logo types

*How to make use of data set and build custom detector using efficient det-d3 pipeline

*This repository has one more dataset

B) Object detection under low illumination

*Objective: to detect objects on the road under low light conditions – fog, haze, rain, etc

* application: This is an important part of autopilot, because it can detect objects, so it is safer under adverse conditions.

*Details: 15K + annotation at 7500 frames on 12 different object types

*How to use data set and efficient det-d3 pipeline to build custom detector

C) Lara traffic light detection data set

*Objective: to detect traffic lights and classify them into red, green and yellow

* application: setting rules for ADAS and automatic driving vehicle system at road network intersections.

*Details: 11K frames and 20K + annotated lights for three traffic types

*How to make use of data set and establish user-defined pipeline detection system using mmdet-faster-rcnn-fpn50

D) Human detection using infrared image

*Objective: to detect people in infrared image

* application: autopilot equipped with infrared camera to detect objects in bad conditions.

*Details: 30 video sequences with 1K + annotation

*How to use data set and MX RCNN pipeline to build custom detector

E) Pothole detection data set

*Objective: to detect potholes from road images

*Application: detect road topography and potholes to achieve smooth driving.

*Details: 700 images with 3K + annotation in potholes

*How to use data set and m-rcnn pipeline to build custom detector

F) Nexet vehicle detection data set

*Objective: to detect road images of vehicles

*Application: vehicle detection is the main part of automatic driving

*Details: 7000 images with 15K + annotation on 6 types of vehicles

*How to use datasets and tensorflow object detection to build custom detector API

G) Bdd100k ADAS data set

*Objective: to detect objects on the road

*Application: detecting vehicles, traffic signs and people is the main component of automatic driving

*Details: 100k image, providing 250K + annotation for 10 types of objects

*How to use datasets and build custom detectors using tensorflow object detection API

H) Linkopings traffic sign data set

*Objective: to detect traffic signs in images

*Application: detecting traffic signs is the first step to understand traffic rules

*Details: 3K image, providing 5K + annotation for more than 40 types of traffic signs

**How to use data set and use mmdet cascade mask RCNN to build custom detector

Fashion, retail and marketing related data sets

A) Billboard detection (secondary sampling openimages data set) data set

*Objective: to detect billboards in images

*Application: detecting billboards is a key part of automatically analyzing marketing activities across the city

*Details: 2K image with 5K + Notes on billboard

*How to use datasets and retinanet to build custom detectors

B) Deepfashion2 fashion element detection data set

*Objective: to detect fashion products, clothing and accessories in images

*Application: application fashion detection has a huge application range from data sorting to recommendation engine

*Details: 490k image, with about 100 annotation object classes

*How to make use of data set to build self defined cornetnet Lite pipeline detector

*Another fashion related data set is the Taobao product data set

C) QMUL openlogo logo detection data set

*Objective: to detect different logos in natural images

*Application: analyzing the frequency of logos in videos and natural scenes is crucial to marketing

*Details: 16K training pictures, including the identification of various brands – food, vehicles, chain restaurants, delivery services, airlines, etc

*How to use data set and MX RCNN pipeline to build custom detector

Sports related data sets

A) Football detection data set (secondary sampling from openimages data set)

*Objective: to detect football across frames in video

*Application: it is very important to detect the football position in the automatic analysis such as offside

*Details: About 3K training images.

*How to use data set and yolo-v3 pipeline to build custom detector

B) Playing card type detection

*Objective: to detect and classify cards in natural images

*Application: the possible application is to analyze the winning probability of different card games

*More than 500 images in 52 card types

*How to make use of data set and build a custom detector using MX RCNN pipeline

C) Football player detection in thermal image

*Goal: use thermal images to locate and track players

*Application: tracking players in the game is a key part of generative analysis

*Details: more than 5K + annotated 3K + images.

*How to use data set and mmdet quick RCNN pipeline to build custom detector

Data sets related to security and monitoring

A) Mio-tcd vehicle detection in CCTV traffic camera

*Objective: to detect vehicles in CCTV cameras

*Application: detection of vehicles in CCTV cameras is a key part of security monitoring applications

*Details: 113k image with 200K + annotation on more than 5 types of vehicles

*How to use data set and mmdet retinanet pipeline to build custom detector

B) Wider personnel detection data set

*Objective: to detect people in CCTV and natural scene images and videos

*Application: personnel detection based on CCTV constitutes the core of security and monitoring applications

*Details: 10K + image and 20K + annotation can detect pedestrians

*How to use datasets and build custom detectors using cornernet Lite pipes

C) Protective equipment – helmet and vest testing

*Objective: helmet and vest of testing personnel

*Application: This is an integral part of security compliance monitoring

*Details: 1.5k + images and 2K + annotations can detect people, helmets and vests

*How to use mmdet cascade RPN with datasets and custom detectors

D) Anomaly detection in video

*Objective: to classify videos according to the operations performed in the videos

*Application: real time detection of anomalies helps prevent crime

*Details: 1K + video corresponding to 10 exception categories.

*How to use datasets and mmaction-tsn50 pipeline to build a custom classifier

Medical image data set

A) Ultrasound brachial plexus (BP) neural segmentation data set

*Objective: to segment some nerve types in ultrasound images

*Application: through the use of indwelling catheter which can block or reduce the source pain, it is helpful to improve the pain management.

*Details: 11K + image and related examples mask, used to detect neural network

*How to use data set and build custom detector

B) Pannuke cancer instance segmentation in cells

*Objective: to segment different cell types in slide images

*Application: automatic analysis of megabyte data in the presence of cancer cells and dead cells

*Details: 3K + image with associated instance mask for detecting different cell types

*How to use datasets and build custom detectors

Satellite imaging data set

A) Road segmentation in satellite image

*Objective: to segment the lane route in satellite image

*Application: help with urban planning and road monitoring

*Details: 1K + image and related instance mask can detect different road areas

*How to use data set and build custom detector

B) Traversable region segmentation in synthetic lunar images

*Objective: to segment rocks and find traversable areas in lunar images

*Application: the basic element of autonomous rover path planning

*Details: 10K + images with related instance masks to detect different rocks and flat ground

*How to use datasets and build custom detectors

C) Vehicle and swimming pool detection in satellite images

*Objective: to detect vehicles and swimming pools in satellite images

*Application: This is a key part of property tax estimation

*Details: 3.5K + pictures, 5K + Notes on cars and swimming pools

*How to build a custom detector using data set and cornernet Lite pipeline

D) Segmentation of road and residential area in aerial image

*Objective: to segment roads and residential areas in satellite images

*Application: This is a key part of property tax estimation

*100 ultra high resolution images with segmentation mask

*How to use datasets and build custom detectors

*Another similar road segmentation dataset and related training code

E) Water segmentation in satellite image

*Objective: to segment water in satellite image

*Application: it is important to understand how water bodies change and evolve over time

-100 ultra high resolution images with segmentation mask

*How to use data set and build custom detector

*Another such data set is deepglobe land cover classification and its related use criteria

Wildlife related data sets

A) Tiger detection dataset (sampled from openimages)

*Objective: to detect tigers in natural and UAV images

*Application: monitoring endangered species

*Details: 2K + image with 4K + annotation.

*How to use datasets and cornernet Lite pipes to build custom detectors

*Another such dataset could be the monkey detection dataset and its related tutorial

B) Zebra and giraffe detection data sets

*Objective: to detect zebra and giraffe species in natural and UAV images

*Application: monitoring endangered species

*Details: 5K + image with 5K + annotation.

*How to use data set and use eficiencydet-d3 pipeline to build custom detector

C) Caltech camera trap dataset

*Objective: to detect animals in trap camera type images

*Application: monitoring endangered species

*Details: 10K + image with 8K + annotation.

*How to use datasets and retinanet pipes to build custom detectors

*Another such camera data set and related training code

D) Elephant detection data set (sampling from coco data set)

*Objective: to detect elephant species in natural and UAV images

*Application: monitoring endangered species

*Details: 5K + image with 5K + annotation.

*How to use data set and mmdet maskrcnn to build custom detector

Underwater data set

A) Turtles found in the wild

*Objective: to detect turtles in underwater images

*Application: monitoring endangered species

*Details: 5K + image with 5K + annotation.

*How to use data set and use effective data to build custom detector

*Similar data sets can be used to monitor underwater fish

Related codes

B) Underwater garbage detection data set

*Objective: to detect marine garbage

*Application: monitoring and control of marine waste

*Details: 2K + image with 5K + annotation.

*How to use data set and use effective data to build custom detector

*More complex pixel based garbage classification data sets and related codes

C) Suim underwater object detection data set

*Objective: to segment underwater objects

*Application: autonomous underwater vehicle path planning, tracking divers and monitoring marine species

*Details: 1.5k + image and 1.5k + annotation mask.

*How to use data set and build custom detector

D) Salty underwater fish recognition data set

*Objective: to detect marine species in underwater images.

*Application: monitoring marine species

*Details: 89 videos to detect fish, crabs, shrimp, jellyfish, starfish

*How to use data set and mmdet to build custom detector — fast RCNN pipeline

Data sets related to text analysis

A) Document layout detection data set

*Objective: to detect document layout for further analysis

*Application: it is necessary to segment the image into different parts, so that the rule-based NLP and text recognition can be further applied.

*Details: 5K + image, tag with 10K + annotation, such as paragraph, image, title.

*How to use data set and MX RCNN to build custom detector

*In the document named iiit-ar-13k, there is a very similar data set for graphic component detection. This is how to use the data set and train the model on it

B) Total text dataset

*Objective: to locate text in natural scenes

*Applications: basic components identified using OCR

*Details: 1.5k + image with 5K + polygon annotation

*How to use data set and text snake pipeline to build custom detector

C) YY MNIST simple OCR data set

*Objective: to locate and classify numbers in white background images

*Applications: basic components identified using OCR

*More than 10 categories of 1K images with 2K + annotations

*How to use datasets and retinanet pipes to build custom detectors

Other data sets

A) Taco garbage detection data set

*Target – locate and segment all kinds of garbage in the image

*Application: the key component of an automatic robot trying to solve the garbage problem in public places

*Details: 15K + annotated 10K images with more than 20 different types of garbage objects

*How to use datasets and retinanet pipes to build custom detectors

B) General object detection data set for indoor scenes

*Objective: to locate and detect indoor objects in the image

*Application: automatic tagging of images in real estate and rental sites with Amenities

*Details: more than 10 different categories of indoor objects (e.g. appliances, beds, curtains, chairs, etc.)

*How to use datasets and retinanet pipes to build custom detectors

C) Egohands hand cut dataset

*Objective: to segment hands in natural scenes

*Application: the first step of understanding gesture, and the application in human-computer interaction and sign language recognition

*Details: 4.8k + image and corresponding hand mask.

*How to use datasets and retinanet pipes to build custom detectors

D) UCF action recognition data set

*Objective: to classify videos according to the operations performed in the videos

*Application: tagging video is very important for storing and retrieving a large number of videos

*Details: 1K + videos corresponding to 101 action categories.

*How to use datasets and mmaction-tsn50 pipeline to build a custom classifier

E) Oil tank data set

*Objective: to detect oil tanks in satellite images

*Application: tracking oil tanks

*Details: 10K + image with 10K + annotation.

*How to use datasets and retinanet pipes to build custom classifiers

Other action recognition data sets

A) Stair action recognition data set and how to train model on it

B) A2D action recognition data set and how to train model on it

C) Kth action recognition data set and how to train the model on it

appendix

For more details about the tutorial, visit our GitHub page

Link to the original text:https://medium.com/towards-artificial-intelligence/50-object-detection-datasets-from-different-industry-domains-1a53342ae13d

Welcome to panchuang AI blog:
http://panchuang.net/

Sklearn machine learning official Chinese document:
http://sklearn123.com/

Welcome to pancreato blog Resource Hub:
http://docs.panchuang.net/