Computer vision and image processing are used widely in sports to significantly influence athletes and team performance. Implementing computer vision techniques is a turning point in the transformation and development of the sports industry. Sports companies use data collected during games to train real-time machine learning and artificial intelligence models for in-game strategies.
This post will discuss the top 7 sports datasets available in Roboflow Universe.
1. Cricket, Football, Baseball Classification
Link: https://universe.roboflow.com/popular-benchmarks/cricket-football-baseball-classification
Project Type: Classification
Subject: Cricket-Football-Baseball
Classes: cricket, baseball, football
Download Formats: YOLOv5, YOLOv7, MT-YOLOv6, COCO JSON, YOLO Darknet, Pascal VOC XML, TFRecord, CreateML JSON, etc
This dataset contains 252 images of playing cricket, football, and baseball (original dataset from Kaggle - Bikram Saha). The images are split into cricket - 95 images, football - 77 images, baseball - 79 images. This is a dataset for image classification in sports. This model will help to identify if the sport or activity occurring in the image or video feed is, or most closely resembles, cricket, football, or baseball.
Test the model's performance by calling Roboflow's API pretrained on the images.
2. Helmet Detecting Model Computer Vision Project
Link: https://universe.roboflow.com/helmet-ytw2w/helmet-detecting-model
Project Type: Object-Detection
Subject: football-helmets
Classes: Helmet, Not_helmet
Download Formats: YOLOv5, YOLOv7, MT-YOLOv6, COCO JSON, YOLO Darknet, Pascal VOC XML, TFRecord, CreateML JSON, etc
The helmet detection model is an object detection dataset with two classes that allows you to detect whether football players are wearing a helmet. The dataset contains 4.3K train, 416 valid, and 209 testing images with auto-orientation and resizing applied to each. This dataset can help detect helmets, track detected helmets, and identify impact to tracked helmets on the field.
Test the model's performance by calling Roboflow's API pretrained on the images.
3. Shuttlecock Computer Vision Project
Link: https://universe.roboflow.com/mathieu-cartron/shuttlecock-cqzy3
Project Type: Object-Detection
Subject: Shuttlecock
Classes: Shuttlecock, Null
Download Formats: YOLOv5, YOLOv7, MT-YOLOv6, COCO JSON, YOLO Darknet, Pascal VOC XML, TFRecord, CreateML JSON, etc
The shuttlecock computer vision project is a dataset containing 5.6K trains, 1.6K valid, and 816 test images of shuttlecocks for an object detection project. The images in the dataset have auto-orientation applied and are resized to 640*640 pixels. This dataset can be used to train object detection models that allow the embedded cameras to track the movement of shuttlecocks during the competitions allowing them to determine the trajectory in 2D and 3D and know when the shuttle hits the ground.
Test the model's performance by calling Roboflow's API pretrained on the images.
4. Golfball Computer Vision Project
Link: https://universe.roboflow.com/anna-gaming/golfball
Project Type: Object-Detection
Subject: ball
Classes: golfball, Null
Download Formats: YOLOv5, YOLOv7, MT-YOLOv6, COCO JSON, YOLO Darknet, Pascal VOC XML, TFRecord, CreateML JSON, etc
The golfball computer vision project is a dataset containing 31K trains, 3.5K valid, and 3.5K testing images of a golf ball from different angles. Each image has auto-orientation applied and is resized to 416*416 pixels using adaptive equalization. The project can be used to train an object detection model capable of detecting golf balls in images and videos. The dataset has excellent applications like sports broadcasting, automated ball tracking, scoring, finding a lost ball, and more.
Test the model's performance by calling Roboflow's API pretrained on the images.
5. Boxpunch Detector Computer Vision Project
Link: https://universe.roboflow.com/markmcquade/boxpunch-detector
Project Type: Object-Detection
Subject: Punches
Classes: bag, cross, hook, jab, lead hook, lead uppercut, no punch, rear hook, rear uppercut, uppercut
Download Formats: YOLOv5, YOLOv7, MT-YOLOv6, COCO JSON, YOLO Darknet, Pascal VOC XML, TFRecord, CreateML JSON, etc
The boxpunch computer vision project is a dataset containing 606 train, 47 valid, and 25 test images of a boxer punching a bag. All the images in the dataset have auto-orientation applied and are resized to 416*416 pixels. This dataset can be used to train a model that can capture different types of punches thrown during boxing matches and training sessions. It can be used to count the total punches thrown in a round, keeping track of intensity, and understanding flaws in punch technique.
Test the model's performance by calling Roboflow's API pretrained on the images.
6. Cyclist Database Computer Vision Project
Link: https://universe.roboflow.com/pawel-brzozowski/cyclists_database
Project Type: Object-Detection
Subject: cyclists
Classes: cyclist_back, cyclist_front, cyclist_side, person
Download Formats: YOLOv5, YOLOv7, MT-YOLOv6, COCO JSON, YOLO Darknet, Pascal VOC XML, TFRecord, CreateML JSON, etc
The cyclist database computer vision project is a dataset containing 4.4K train, 572 valid, and 190 test images of people riding bicycles from multiple directions (front, back, and side). The images in the dataset have auto-orientation applied. This dataset has multiple use cases, from training self-driving cars to be cautious of cyclists to sports broadcasting to gather information about how cyclists are positioned.
Test the model's performance by calling Roboflow's API pretrained on the images.
7. Football Players Computer Vision Project
Link: https://universe.roboflow.com/bronkscottema/football-players-zm06l
Project Type: Object-Detection
Subject: american-football-players
Classes: american-football-players, center, db, lb, qb, skill
Download Formats: YOLOv5, YOLOv7, MT-YOLOv6, COCO JSON, YOLO Darknet, Pascal VOC XML, TFRecord, CreateML JSON, etc
The football player computer vision project is a dataset containing 1.8K train, 100 valid, and 50 test images of football players in straightforward positions like center, quarterback (qb), defensive back (db), and linebacker (lb). All the images in the dataset have auto-orientation applied. This dataset can detect players across various scenes in real-time and help identify the specif plaers. In advanced use cases, if trained with other models, this dataset can assist with referee rule enforcement and regulations during a game.
Test the model's performance by calling Roboflow's API pretrained on the images.
Using Open Source Sports Datasets for Computer Vision
Computer vision applications in sports are emerging all over the globe, covering post-game analysis, in-game activities, sponsor opportunities and much more. Today's article provided some insight into datasets you can use for your sports applications.
Sign up for a free Roboflow account to add images from open source datasets and train a computer vision model in one click. You'll have a fully running computer vision pipeline in no time.