Datasets:
PlantDoc-corrected: 2565 images / 8887 boxes / 29 Title-Case classes. Removed 12 train-test leak duplicates, 11 empty XMLs, 11 unlabelled images and 4 out-of-canvas annotations; normalised EXIF orientation on 35 images; sanitised 234 URL-tainted filenames (25 extension-less, 12 script extensions); fixed 4 zero-size VOC canvases; rebuilt seeded 10% val split from official TRAIN
a89f439 File too large to display, you can check the raw version instead.