Free tool
Dataset Health Report
Check one annotation file for class imbalance, crowded images, near-duplicate boxes, and out-of-bounds boxes before you train. Add two or three COCO split files to find images that appear in more than one split.
Loading report…
Frequently asked questions
- What does the dataset health report check?
- It summarizes class balance, annotation density per image, and box geometry. It flags a dominant class, sparse classes, crowded images, near-duplicate boxes, boxes outside the image, duplicate file names, and listed images with no annotations.
- Which files can I analyze?
- Upload one COCO JSON, CVAT XML, JSONL, CSV, or TSV file of up to 25 MB. The format is detected automatically.
- Can it find images that leak between train and test splits?
- Yes. Upload two or three COCO JSON split files, and the report lists file names that appear in more than one split. It compares file names, not image content.
- Are my files uploaded?
- The files you select are sent to LabelOp servers for processing, and the result comes back as a download. You do not need an account.