100-Class Animal Segmentation Dataset

However, fine-grained segmentation datasets for animals are still very scarce. To this end, we constructed the A100 Seg dataset, which includes 100 common and rare animals, each providing 100-150 images (totaling over 11000 images).

We used a combination of SAM-ViT-B automatic pre segmentation and ISAT manual fine correction to generate high-precision pixel level annotations. Each image contains only one animal, ensuring a clean scene, and providing both ISAT raw JSON (polygon outline) and PNG binary mask formats. A100 Seg is currently the only publicly available dataset that simultaneously meets the criteria of "100 or more animals, ≥ 100 images per category, pixel level segmentation annotation", filling an important gap in this field.