Lab: Compare Localization Overlap

Mission

Run the local Localization Overlap Lab. Compare tight, shifted or under-segmented, and oversized predictions for both boxes and masks. Export a Localization Evidence Record.

Guided start

1. Select Bounding box and Shifted / under.
2. Run the measurement.
3. Preserve intersection 12, predicted area 16, target area 16, union 20, and IoU 0.600.
4. Switch only the prediction to Oversized and rerun.
5. Explain why covering the whole target still produces a lower score.

Procedure

1. Run all three bounding-box predictions.
2. Record every intersection, area, union, and IoU.
3. Switch to segmentation masks and run all three cases.
4. Compare an under-segmented mask with an oversized mask.
5. Preserve your first localization claim and write a bounded correction.
6. Complete Observation, Measurement, Decision, and Limitation.
7. Export the record and confirm it names ai-for-beginners-localization-overlap-lab.

Success criteria and boundary

• at least three IoU values are independently reproducible;
• box and mask output types remain distinct;
• shifted, missed, and extra regions are named accurately;
• the corrected claim follows only supplied geometric evidence; and
• the limitation rejects unsupported claims about class, annotation quality, safety, fairness, or other images.

Use only supplied fictional annotations. The Lab activates no camera, identifies no person, and supports no surveillance or safety decision.