Accurate Balls Detection

Introduction

The dataset includes, 1318 grayscale images and contains highly accuratelyannotated balls. The annotations are available in COCO format. This datasetcan be used to train models to detect any circle-shaped objects, e.g.calibration targets. The dataset is split into:

train: 1120 images

valid: 128 images

test: 76 images

Processing and Augmentation

The following augmentation was applied to create 2 versions of each source image:

Random exposure adjustment of between -10 and +10 percent

Random Gaussian blur of between 0 and 1.5 pixels

Salt and pepper noise was applied to 3 percent of pixels

References

This dataset is a subset of the following datasets:

kaggle-sports-balls-multiclass-image-classification

roboflow-ball-finder

Custom recorded images of an exercise ball