OpenTrace ML: Python toolkit for road and route intelligence

General talk
1

Hi everyone,

I’m developing OpenTrace ML, an Apache-2.0 Python library connecting:

  • road-damage computer-vision detections;
  • incremental traffic forecasting;
  • GPS traces and GeoJSON;
  • transparent route-reliability scoring.

The current stage includes model-agnostic detector adapters, detection and forecasting metrics, rolling backtests, and examples using RDD2022-style annotations, UCI traffic data, and OpenStreetMap.

OpenStreetMap is used as externally licensed network data. The library does not bundle OSM extracts or automatically submit edits to OpenStreetMap.

I would especially appreciate community feedback on:

  1. Creating a small, reproducible offline OSM fixture for tests.
  2. Choosing useful routing or map-matching integrations.
  3. Handling OSM attribution and derived-data licensing correctly.
  4. Designing a responsible human-review workflow for any future ML-generated road suggestions.

Repository: GitHub - vrajpatell/opentrace-ml · GitHub

Offline OSM fixture issue: Add a tiny offline OpenStreetMap integration fixture · Issue #3 · vrajpatell/opentrace-ml · GitHub

Contributions and technical feedback are welcome.

2

Hello Vraj

I can’t help you with any of the programming details, but I have a question of my own.

Where are you getting the road-damage date from?

3

Hi Graeme, thanks for asking.

At the moment, the reference road-damage data is the public RDD2022 dataset, available on Figshare:

https://figshare.com/articles/dataset/21431547

It provides annotated road-crack and pothole images and is listed under CC BY 4.0. OpenTrace ML does not bundle the dataset or model weights—the example reads a user-downloaded RDD2022 directory, while the automated tests use a small original synthetic XML fixture so CI does not redistribute third-party data.

The source and licensing notes are documented here:

For future real-world observations, the intention is to use explicitly consented data with privacy safeguards, not scraped imagery or OpenStreetMap as the source of road-damage detections.

4

Thanks.

IANAL, but I think that’s possibly now going to be a tricky one, as we need a waiver to be able to use CC BY 4.0 data in OSM: Use of CC BY 4.0 licensed data in OpenStreetMap | OpenStreetMap Blog

Anybody else care to comment?

5

Maybe I am missing something, but I don’t see any mention of using this data in OSM? I think we are tallking about an application that uses OSM data among other sources, not one for contributing to OSM?

1 Like
6

Thanks, Alan—that’s correct. RDD2022 is used only for application-side experiments. OpenTrace uses OSM as the base network and does not upload RDD2022-derived detections to OSM.

Graeme’s licensing point is important. Any future OSM contribution workflow will require an authorized data source, confirmed licence compatibility, and human review. Until then, generated results will remain a separate OpenTrace layer.

OSM, routing, GIS, and Python contributors are welcome to review the project and help with the offline OSM fixture: GitHub - vrajpatell/opentrace-ml · GitHub

1 Like
7

Thank you - I misunderstood.