Prompt-Based Video Segmentation & Tracking

A pipeline that takes a point or box prompt on the first frame and carries the resulting mask through a video. SAM handles the segmentation, YOLOv8 provides detection and identity across frames, and the two are reconciled so a mask survives occlusion and re-entry.

Includes a comparison report on where each component fails — SAM's prompt sensitivity, YOLO's identity switches — and how the combination covers for both.