Sports balls - multiclass image classification

Over **9,000 images** of sports balls from **15 different sports**. These are: american football, baseball, basketball, billiard ball, bowling ball, cricket ball, football, golf ball, field hockey ball, hockey puck, rugby ball, shuttlecock, table tennis ball, tennis ball and volleyball.

Images were **scraped from google images** and then **manually reviewed** to remove misclassified images (although I might have missed some). Duplicates were removed using a dHash style algorithm. The **train/test split is 80/20**.

This is an interesting dataset because **several images are misleading**, e.g. some balls have been painted to look like other balls. See some examples below:

<img src="https://i.postimg.cc/jdHLWfbs/golf-ball-681-copy.jpg">

<img src="https://i.postimg.cc/85vT28k1/cricket-ball-1156.jpg">

<img src="https://i.postimg.cc/yYs7brMH/american-football-58.jpg">

本数据集包含来自15种不同运动项目的9000余张体育球类图像。具体包括：美式足球、棒球、篮球、台球、保龄球、板球、足球、高尔夫球、曲棍球、冰球、橄榄球、羽毛球、乒乓球、网球和排球。图像源自谷歌图片，并经过人工审核以去除误分类的图像（尽管可能遗漏了一些）。通过dHash风格算法移除了重复图像。数据集的划分比例为80%的训练集与20%的测试集。该数据集颇具研究价值，因为其中部分图像存在误导性，例如某些球类被涂漆以模仿其他球类。以下为部分示例：




搜集汇总

背景与挑战

背景概述

该数据集是一个多类图像分类数据集，包含15种运动球类的9,000多张图像，图像经过人工审核和去重处理，采用80/20的训练测试划分，并包含一些具有挑战性的误导性图像。