meaculpitt/ScoreVision-Numberplate

Single-element chute for the SN44 element manak0/Detect-number-plates-1-0 (activates 2026-04-12T12:02 UTC).

• Architecture: YOLO11s exported to ONNX (≤30 MB hard cap)
• Input: 1408×768 RGB frames (the validator's native resolution)
• Output: bounding boxes around license plates (single class numberplate, absolute-pixel xyxy coordinates)
• Inference: ONNX Runtime CUDA EP, Gaussian Soft-NMS (sigma=0.5) for plate-dense scenes
• Latency target: p95 ≤ 50 ms end-to-end

Trained on a 344k-image combined corpus of:

• Roboflow license-plate-recognition-rxg4e v4
• Roboflow vehicle-registration-plates-trudk v2
• Roboflow car-license-fj1kd v4
• helpme-8ixem/anpr-iuzao (real CCTV ANPR footage)
• CCPD2019 (Chinese City Parking Dataset, ~200k overhead surveillance frames)

Sibling repos in this org

• meaculpitt/ScoreVision-Vehicle — vehicle (manak0/Detect-detect-vehicle), served by the insect chute
• meaculpitt/ScoreVision-Petrol — petrol station components

This repo (ScoreVision-Numberplate) is intentionally a separate single-element submission deployed under its own chute slug, not part of the unified insect chute.