Datasets:

jjjlimaus
/

person-new-cctv-sam3

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.

Person New CCTV — SAM3 Annotated

YOLO-format person boxes for synthetic CCTV images from jjjlimaus/person-new-cctv-synthetic.

Annotator: SAM3 (prepare_dataset_with_sam3.py --mode person)
Class: person only (nc: 1)
Augmentation: original images plus horizontal flips (_aug_hflip), each flip re-annotated with SAM3
Samples: 2060 (train 1648 / val 412 / test 0)
Split: 8:2:0 (seed 42); flips stay in the same split as the source image
Resolution: all images stretched to 1024×1024 (original 1376×768; aspect ratio not preserved). YOLO boxes remapped in pixel space to the new canvas.
Layout: Ultralytics {train,val}/{images,labels}/ + dataset.yaml

Access: public + gated (manual).