You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

LocAny annotations

Original annotations with saved Rex-Omni, Qwen, YOLO-E and SAM3 predictions where available. YOLO-E and SAM3 masks are stored as COCO RLE alongside their boxes. Images use the original media references.

49/49 original views uploaded (11,770,115 image records).

Files: source/datasets/<dataset>/views/<view>/records.jsonl. Original fields are unchanged; _annotations holds model outputs. _cleaned_strict siblings contain the existing filtered annotations. These do not apply a new union of judges.

See source/manifests/views.jsonl for counts and hashes, and runs.json for the saved prediction snapshot. Missing predictions are marked not_available. The ongoing Qwen pointing-to-box run includes only chunks completed at snapshot time.

Oversized JSONL files are stored as records.jsonl.gz. Run gzip -dk records.jsonl.gz before using their existing byte-offset index and metadataset. The manifest records compressed and original SHA256 values.

Downloads last month
55