Documentation menu

Documentation

Build and run a workflow

On this page

A workflow connects typed node outputs to compatible inputs. Editing a graph, saving a revision, running on fixed images and applying to a device are separate operations.

Browser Lab workflow

  1. Choose an example or import your image. Inspect its dimensions and whether the Lab scaled it.
  2. Select a node to inspect its parameters. Right-click the canvas to add a supported node. The public Lab search lists executable browser nodes and hides camera, cloud API and excluded candidate nodes.
  3. Connect compatible ports. A cycle, unconnected required input or incompatible type is refused with a reason.
  4. Apply the parameter edit or discard it. A pending edit blocks execution so the displayed parameters and executed graph remain consistent.
  5. Select Run, then inspect Results. Use Trial selected node for a bounded local diagnostic; check its scope before interpreting the result.
  6. Export the workflow and development result from Workflows and samples if you want a durable copy.

Implemented browser nodes include crop/pad, grayscale, threshold, morphology, integer gray filters, rotation, blob location, numeric/logic operations and selected classic/rule/barcode processing. Some nodes require an explicit Gray8 conversion. A node preserved in an imported graph can still be unsupported by this runtime; there is no automatic fallback to a camera, cloud engine or PC service.

Blob location selects the largest eligible 8-connected Gray8 component. It reports original-image center, angle, area and eligible-component count. Nearly round objects have no useful orientation; the locator has no confidence score. Missing pose values remain missing, and comparisons on missing values evaluate false.

For physical measurement, use only a reviewed calibration, an appropriate verified plane and the declared units. An invalid/out-of-plane measurement requires review; a pixel coordinate is not automatically a calibrated length.

Windows PC workflow

Start an offline project from the workbench with a fixed set of images and known answers. Choose the available template and inspect the graph. Add/connect nodes, apply parameters, and review the number of unsubmitted changes.

Use Validate and submit to save a workflow revision. Confirm the revision and changes in the dialog. Submission saves the proposal and does not execute it. Then choose Run current revision to evaluate the fixed images. This consumes the task's PC trial budget; it is not a new camera exposure or a camera-parameter write.

Inspect which workflow version, input and task produced the result. If another client changed the workflow, your old draft must not overwrite the current revision. Read the latest version and human review before deciding how to proceed.

Compare iterations

Compare the decoded content or measured value against independently known answers, the failure/review cases, and elapsed time. A visually cleaner image can still produce the wrong decision.

Keep development images separate from a frozen validation set. Derived crops, repeated runs and augmented copies of the same physical item are not independent new samples. Define acceptance thresholds before inspecting validation results.

Save and persistence

A public Lab export is a local file download. Session drafts/history can be lost when closing a tab and do not create an owner-private server account. Account workspace, private revision history and own-Codex relay features belong to separate research/candidate builds, not the public Lab.

Desktop workflow revisions and run receipts belong to the task service's data root. Preserve that root, workflow records and evidence together when backing up.

Engineering preview. Validate the supplied release and site acceptance before production use.

Cognex is a third-party trademark. No affiliation or certification is implied.

Questions about your own setup? Contact Deyi Vision

Contact

Direct RFQ contact

Talk to engineering about the inspection problem.

Send sample images, competitor model, FOV, working distance and line speed before model selection.

Target: selection brief within 24h
Send sample images