Velospatial

Your dashcam can teach a road model.

We are training an Australian-owned model that reads road damage : potholes, edge break, faded line marking : from ordinary driving footage. Contributed frames become part of its training data.

Your video never leaves your computer

This page pulls still frames out of your video in your browser : one every couple of seconds : and uploads only those stills. The video file itself is never sent to us.

Contribute

Used only to record your agreement to the terms and for questions about your contribution. Not marketing, not shared.

Contributor terms - v1, 29 August 2026

Plain-English summary: you are giving these frames away, permanently. Read the whole thing; it is short.

1. What you are giving us

The still frames this page extracts from your footage. You assign to Velospatial all rights in those frames, including copyright, worldwide and permanently. To the extent any right cannot be assigned, you grant Velospatial a perpetual, irrevocable, worldwide, royalty-free, transferable licence to use, store, modify, publish excerpts of, and train machine-learning models on them.

2. No take-backs

Once submitted, frames become part of a training dataset and derived models and cannot be recalled, withdrawn or deleted on request. Do not contribute anything you might ever want back. Velospatial may remove material at its own discretion but is not obliged to.

3. What you promise

You are 18 or older. You recorded the footage yourself, or hold every right needed to give it to us on these terms. It was recorded lawfully, from a vehicle you were entitled to record from. It does not include audio (this page never extracts audio), footage of identifiable private situations, or anything depicting an injury.

4. People and plates in shot

Public-road footage incidentally captures number plates and passers-by. We use frames only for training and evaluation, restrict access to them, and blur identifying details in anything we publish. If you appear identifiably in your own footage, submitting is your consent to this use.

5. No payment, no ownership of the model

Contributions are voluntary and unpaid. Models trained on the dataset, and everything they produce, belong to Velospatial alone.

6. The boring but real bits

These terms are governed by the laws of Victoria, Australia. If part of them is unenforceable the rest still stands. Questions or concerns about something you submitted: contact us through the site.

Drop dashcam videos here or click to choose
MP4 or MOV · up to 5 files · frames extracted locally

Daily limit 600 frames per contributor a day. Night, rain and blur are all useful : a model must learn what bad conditions look like too.