Tool

Web Labeler

The corner-labeling web GUI — the labeling-party tool. Part of Step 2 of the plan.

Run it

cd /home/ubuntu/spar
source venv/bin/activate
python web_labeler/labeler_server.py --images data/raw/ --labels data/labels/ --port 5050

Everyone on the same network opens http://<host-ip>:5050 in their phone browser. Each person enters their name; images auto-assign so nobody labels the same plate twice. Click the 4 plate corners, type the plate text, save. Labels land as JSON in data/labels/.

Controls — Web GUI

Input	Action
Click / tap	Place corner (in order)
u or Undo button	Undo last corner
Reset button	Clear all corners
n / Enter or Save button	Save + next image
Skip button	Skip to next image

Controls — Desktop tool

Key	Action
Left-click	Place corner (in order)
u	Undo last corner
r	Reset all corners on current image
n / Enter	Save + next image
q	Quit (saves progress)
+ / -	Zoom in/out

Verify labels

python check_labels.py -i data/raw/ -l data/labels/

Output: JSON files in data/labels/ — one per image. Run the checker to validate and produce a verification collage.

Improvement roadmap

From LABELER_PLAN.md — the labeler's future, in priority order:

1. ✓ Plate text entry — after 4 corners are placed, show a text input for the plate text before saving. Corner flow → text input slides up → confirm → single save with corners + plate_text. Done.
2. Corner editing (drag to adjust) — after all 4 corners are placed, tap-drag individual corner dots to fine-tune. Eliminates undo-reset-retry.
3. Winding order validation — warn if corners aren't in proper TL→TR→BR→BL order; flag self-intersecting quadrilaterals with a ⚠️ badge.
4. Live rectified preview — small rectified view of the plate in the corner of the screen once 4 corners are down. "That doesn't look like WGE7330" → fix it immediately.
5. Review mode — browse finished labels with polygon overlays. Edit or reject bad ones. Filter by labeler.
6. Team presence (WebSocket) — replace 5-second polling with WebSocket for live progress and team visibility.
7. PWA / offline — service worker + manifest. "Add to Home Screen" → standalone app. Queue labels offline, sync on reconnect.
8. Multi-plate support — some images have multiple plates (front + back, two cars in frame). Allow labeling 1–N plates per image.
9. Production server — swap the Flask dev server for waitress or gunicorn. One-line change.