Edge data collection

Everything starts with data. This guide shows how AcuSight collects images from your edge devices in real time, keeps that data diverse, and hands it off to a project — the first step toward training a model.

​

How collection works

Once a device is provisioned, it captures images and streams them to AcuSight continuously — no manual upload, no copying files off an SD card. New images arrive on their own: they show up under the device’s Edge Data, and the device appears on Home with live status.

​

Batches

Incoming images are grouped automatically into batches. A batch stays collecting as images arrive, then auto-completes once it’s full or the device goes quiet — no manual bookkeeping. Each completed batch is a tidy unit of data you can move into a project.

​

Keeping your data diverse

A model only learns from variety. Hundreds of near-identical frames of the same still object teach it very little — and can quietly bias it toward one view. To prevent that, the AcuSight edge agent filters frames on the device, before they’re uploaded: it measures the difference between successive frames and automatically drops ones that are too similar — where little has changed since the last. It happens on its own, with nothing to configure, so each batch stays visually diverse and representative of what the device actually sees. The result is a smaller, richer dataset that trains a better model.

​

Move it into a project

Collected data waits in the raw pool until you put it to work. When you’re ready:

1. Open the batch under Data.
2. Select Move Batch to Project and choose your project.

The batch lands in your project’s Annotate board, ready to label.

​

Next steps

Collection is the first stop in the data workflow — from here, the rest of this section takes your images to a training-ready dataset: