DIS5K

DIS5K (Dichotomous Image Segmentation 5K) High-accuracy dichotomous image segmentation dataset, mirrored to the nobg org for convenience. Each example is an RGB image paired with a high-quality binary segmentation label. image: RGB source image label: binary segmentation ground-truth mask image name: original image file name label name: original label file name Splits: DIS TR (train, 3000), DIS VD (validation, 470), and the four test sets DIS TE1–DIS TE4 (500 each, increasing… See the full description on the dataset page:

Source: Hugging Face Hub (nobg/DIS5K). Metadata imported from the dataset’s Hub tags.