Validate computer vision datasets before training. Identify corrupted images, annotation errors, duplicate samples, potential data leakage, and class imbalance—with actionable recommendations to improve dataset quality.
Detect corrupted images, missing annotations, invalid coordinates, malformed labels, and other dataset integrity issues.
Identify duplicate and near-duplicate images, cross-split inconsistencies, and potential data leakage.
Assess class distributions, background-heavy datasets, and small-object prevalence at training resolution.
Convert supported annotation formats, create balanced dataset splits, and export validation reports.
YOLO, COCO, Pascal VOC, LabelMe, CVAT, KITTI and CreateML, plus YOLO-seg and COCO polygon/RLE workflows.
Indexed, grayscale and RGB masks, including common VOC, Cityscapes, ADE20K and CamVid-style layouts.
Dataset analysis runs locally in your browser. Your files are not uploaded to our servers. Free access, with no account or email required.