Browser vision lab

Build and inspect an image-processing workflow in your browser. Image computation stays on your device. No email address is required to use the lab. The optional contact form only sends the details you choose to provide.

Local image tools

Image tools include grayscale conversion, thresholding, blur, median filtering, normalization, local thresholding, morphology, cropping and padding. Use numerical comparisons, range checks and logic nodes to turn intermediate values into a workflow decision. Inspect each step rather than treating a final pass/fail label as an explanation.

Illustrative scene workflows

All 12 scene cards open their illustrative workflows directly from the workbench. They use synthetic teaching inputs, not customer photographs, completed inspection projects or production performance tests. Dimension and surface-difference workflows require you to prepare, inspect and explicitly confirm their example materials before running:

• Presence checks: illustrate whether a part is present.
• Counting and missing parts: compare a component count with an expected count.
• Position and offset: inspect a part's displacement from a reference position.
• Dimension and tolerance: prepare and inspect the example calibration, then explicitly confirm it. Physical measurements require human-reviewed calibration for the current input.
• Surface difference: inspect the input and independent example reference, then explicitly confirm the material. Run the comparison only with a reviewed reference image.
• OCR and labels: run the local OCR model and inspect the text and confidence it actually recognizes before evaluating the example label rule.
• Barcode traceability: try decoding the example barcode.
• SKU and color: compare a sample color with a target.
• Fill and liquid level: inspect the example fill proportion.
• Assembly and orientation: examine relative component positions.
• Packaging and kits: count the items in an illustrative kit.
• Seals and adhesive beads: inspect example bead widths and breaks.

Opening S04 or S05 does not confirm materials or start a run. Their example materials cover only the fixed synthetic inputs; your own images require separately prepared materials. Changing an input or its mapping, restoring a workflow, or reaching the 24-hour review expiry requires preparation and explicit review again. Confirmation and execution are separate actions.

OCR and reference resources are downloaded from this site only when a workflow needs them; opening the lab or running other scenes does not download them. The first OCR run downloads approximately 35 MiB of model and runtime resources. Recognition runs in your browser; loading or recognition failure is shown as an error, without substituting expected text.

Before using your own image

Choose a browser-decodable image such as PNG, JPEG or WebP, up to 24 MiB. Photos may be reoriented using orientation metadata; images above 4 Mi pixels are downscaled for the working image. Keep the original file and check the displayed dimensions and orientation before interpreting pixel measurements. Workflow debugging processes chosen image files. The separate production workspace connects directly to compatible cameras through the Cognex Web SDK when you explicitly connect.

Run and inspect a trial

1. Open the workbench directly, without submitting contact details.
2. Choose a scene or a local image and adjust the workflow nodes. For S04 and S05, prepare and explicitly review the fixed example materials before running; do not apply them to an unrelated image.
3. Inspect the results on your own samples. A browser trial does not prove camera compatibility or production inspection performance.

If loading does not finish, read the machine vision software guide for requirements and available workflows.