Dataset Health Check (Image Classification)

Image Classification

Overview Tab

High-level summary: total images, annotated count, issues found, duplicates or corrupted files detected.

Quality Checks Tab

Check	What it detects
Small boxes	Bounding boxes smaller than 0.5% of image area
Low resolution	Images below 640×480 px
Box count anomalies	Unusually high or low bounding box count per image
Missing class labels	Class IDs not present in classes.txt
Class imbalance	Classes with significantly more/fewer examples

Actions Tab

• Duplicate image management — side-by-side previews of detected duplicates
• Unannotated image management — lists images with no label file
• Soft-delete — marks images as excluded from training without deleting from disk