Reported Bike Lane and Crosswalk Detector
This repository contains the Reported YOLOv8 segmentation ONNX export used for complaint pre-categorization in civic traffic reporting workflows.
The model artifact is:
reported-v13-optimized.onnx
Model Details
- Format: ONNX
- Architecture: YOLOv8s segmentation
- Input size used by the local postprocessor: 512 x 512
- Source metadata:
Ultralytics YOLOv8s-seg model trained on /content/Reported-13/data.yaml - SHA-256:
64ff4e935369defa74e30fcc4c19474f37f418cf6b6c76a1560d7e9570442879
Classes
The embedded class metadata lists 13 classes:
| Index | Label |
|---|---|
| 0 | blocked bike lane |
| 1 | blocked crosswalk |
| 2 | bullbars |
| 3 | ct |
| 4 | fl |
| 5 | md |
| 6 | medallion |
| 7 | nc |
| 8 | nj |
| 9 | ny |
| 10 | nypd |
| 11 | pa |
| 12 | tlc |
Intended Use
Use this model to detect and segment street-safety complaint cues, especially blocked bike lanes and blocked crosswalks, in images or video frames. The repository includes labels.json for downstream apps that need the class index mapping.
Limitations
This model should be used as an assistive signal, not as the sole source of truth for enforcement or other high-impact decisions. Users should review detections and confirm report details before submitting them to a city or agency.
Postprocessing Notes
The Reported app repository includes a YOLOv8 segmentation postprocessor for this ONNX export. It treats class 0 as blocked bike lane, applies confidence and IoU thresholds, and decodes mask coefficients from the model outputs.
- Downloads last month
- -