Feature request:
Filter images by bounding box size
I’d like to request a dataset feature that allows filtering images based on bounding box size.
Why this is useful
Quickly find images with small objects (hard cases) vs large objects for targeted review.
Identify labeling issues (e.g., tiny boxes).
Build curated subsets (e.g., “small-object training set”) without exporting and scripting externally.
Requested filter options
Filter by bbox area (in pixels² and as % of image area).
Optional: filter by bbox width/height (px), and aspect ratio.
Support per-image aggregation, e.g.:
- Has any bbox smaller than X
- All bboxes larger than X
- Median bbox area in image between A and B