Version 20.05, the latest update to MVTec’s HALCON deep learning software package, features an advanced decoding algorithm for barcode reading that allows codes with bars smaller than one pixel to be scanned. The new release also allows deep learning training on central processing units (CPU) instead of graphics processing units (GPU), enabling standard industrial PCs without dedicated GPUs to be used for deep learning applications.
Boxes of different shapes and sizes can more reliably be detected in version 20.05 and Grad-CAM-based (Gradient-weighted Class Activation Mapping) heatmaps can be calculated on CPUs, without significant speed drops, according to the manufacturer. Anomaly detection has been improved, network training speeds are 10 times faster, inference speed has improved, and trained networks now require less memory and disk space, an improvement aimed at embedded devices.
Finally, surface-based 3D matching has been improved with version 20.05, such that features such as small holes can now be used to improve accuracy and robustness of inspection routines.
GET PRICING
To Learn More:
Contact: MVTec
Headquarters: Munich, Germany
Product: HALCON ver. 20.05
Key Features: Increased support for CPU-based and embedded platforms, improved box detection and surface-based 3D matching, increased training speeds and optimized memory/disk space requirements.
What MVTec says: View more information on HALCON version 20.05.
Share your vision-related news by contacting Dennis Scimeca, Associate Editor, Vision Systems Design
SUBSCRIBE TO OUR NEWSLETTERS
