These are the dataset splits for the paper SOS: Synthetic Object Segments Improve Detection, Segmentation, and Grounding. The SOS-GC-1M collection contains one million samples, each with:
Segmentation masks
Bounding boxes
Referring expressions
All annotations are provided in COCO format.
Clone or download the entire repository.
from huggingface_hub import snapshot_download
snapshot_download(… See the full description on the dataset page:
https://huggingface.co/datasets/weikaih/SOC-GC-1M.
