Image-Video Bounding Box Annotation

I need a clean, production-ready dataset of bounding-box annotations applied to a mix of still images and short video clips. The media is already stored in an S3 bucket, and once we begin I will share a concise taxonomy of classes—think vehicles, people, animals, and any additional categories we finalise together. The task is straightforward but must be precise: draw tight boxes around every instance of each class, keep labels consistent across consecutive frames, and avoid drift in long sequences. You may work in CVAT, Labelbox, VGG Image Annotator, or any other tool that can export COCO JSON or YOLO-format text files; direct integration with AWS SageMaker Ground Truth is welcome. Deliverables • Complete annotation files (COCO JSON or YOLO txt) for all images and extracted video frames • A brief quality-control report describing checks performed (IoU thresholds, peer review, etc.) • A sample export demonstrating correct label structure before full hand-off Acceptance criteria • Every target object is fully enclosed—no clipping, no missed instances • Labels match the agreed taxonomy exactly, one label per object • Output files pass standard COCO/YOLO validators without errors Include a note on your expected daily throughput and any previous projects that highlight your accuracy with bounding-box work. I’m prioritising this format, but if you also handle OCR or more advanced video annotation workflows, feel free to mention it—those skills may prove useful in later phases.

Project ID: 40327900