#insightvisionsystems #deeplearning #cognexvisionsystems
Cognex vision systems make setting up and deploying deep learning applications quick and easy, especially with our In-Sight ViDi. Learn step-by-step how to develop a robust defect detection application using an In-Sight D900 Vision System.
--
ABOUT US:
For over 40 years, Cognex has been making advanced machine vision easy, paving the way for manufacturing and distribution companies to get faster, smarter, and more efficient through automation.
Innovative technology in our 2D and 3D vision systems and industrial barcode scanners solves critical manufacturing and distribution challenges, providing unparalleled performance for industries from automotive to logistics to packaged goods.
Cognex makes these tools more capable and easier to deploy and thanks to a longstanding focus on AI, helping factories and warehouses improve quality and maximize efficiency without needing highly technical expertise.
We are headquartered near Boston, USA, with locations in over 30 countries and more than 30,000 customers worldwide.
CONTACT US:
LinkedIn: / cognex-corporation
--
SUMMARY:
The video demonstrates configuring a deep learning defect detection system for textile inspection. The process involves collecting images of good and defective parts, labeling them within a specialized editor, and training a model to distinguish between them. Key steps include setting the inspection region, defining feature sizes, and applying rotation perturbations to handle varying orientations. After training, the system generates scores and highlights defects. Data extraction tools populate spreadsheet cells with metrics like area and location. Finally, the solution is deployed via a web-based Human-Machine Interface (HMI), allowing users to view live results, adjust thresholds, and integrate outputs with external controllers like PLCs.
KEY TAKEAWAYS:
-Setup includes a camera, GPU-enabled computer, and security dongle for running deep learning tools.
-Collect 15–20 images of good parts and multiple defect types, rotating to cover all angles.
-Rename frames by prepending "good" or "bad" to streamline the labeling process in the editor.
-Drag the Vidi Detect tool into the spreadsheet, set the inspection region, and import labeled images.
-Adjust feature size using a bubble tool on the smallest defect and set a 50% training/test split.
-Enable rotation perturbation to synthetically generate rotated images for robust model training.
-Train the model; it learns to score images, separating good (low score) from bad (high score) parts.
-Use "Insert Getters" to pull defect data (area, perimeter, location) from the tool into spreadsheet cells.
-Publish scores and graphics tags to a default web HMI for real-time monitoring and user interaction.
-Assign a watch cell to the live film strip and customize the HMI for production deployment.