Two angles of the same intersection. Sync them, detect/track people & cars, and ask free-text questions about what's in view.
Not yet auto-synced. Auto-detect gives a starting point from ambient audio — always verify visually and fine-tune above.
Needs both videos processed offline with tracking on (Demo → Tracked) first. Matches tracks by appearance across the sync offset -- a heuristic, not ground truth. Linked objects get a shared "G#" id and color in both views.
Uses the frame + detections at the moment you ask (pause/scrub to point at a specific instant — imitates "now" on a live feed).