Dataset Health Check

Assessing and improving the quality of your dataset.

Follow the Guided Tutorial

The best way to understand Roboflow's dataset Health Check is by following the Dataset Health Check Guided Tutorial, which walks through a health check on a public dataset of hard hat construction workers.

Breaking Down the Health Check (Object Detection)

Understanding your Dataset Health Check helps you create informed decisions about preprocessing and augmentation decisions for your dataset.

The Basics

Images includes the number of images and missing or null annotations. Missing annotations are images that do not have an accompanying annotation file. Null annotations are images that deliberately do not contain any objects. See more on the difference between null and missing annotations from our blog.

Annotations describes the total number of objects annotated (i.e. the number of bounding boxes).

Average Image Size is the size of images in megapixels.

Class Balance

Class Balance shows class (im)balance. Unbalanced data can yield unfavorable results, especially when measuring models with accuracy.