Matrox Imaging adds deep learning capabilities to SDK with two new updates
Matrox Imaging has announced a major update to its Matrox Imaging Library software development kit (SDK), the MIL X service pack, that expands the library’s deep learning capabilities.
Specifically, the MIL X update expands the software’s classification tools and allows for the training of convolutional neural networks to analyze images of highly textured, naturally varying, and acceptably deformed objects. The update supports training dataset creation, including image labeling and augmenting with synthesized images, and adds the ability to monitor and analyzing the training process. Pre-defined CNN architectures can also be trained via transfer learning and fine-tuning.
The image-oriented classification tools in the MIL X update include a global approach that assigns images to classes and a coarse segmentation approach that maps image neighborhoods according to categories, identifying and roughly locating the presence of specific features or defects. MIL X also expands the registration toolset with a new high dynamic range (HDR) imaging feature that produces an image with larger dynamic range than possible in conventional images.
A companion update for the MIL CoPilot interactive environment also supports deep neural network training by allowing users to label and augment datasets, visually monitor training, and see results in tables.
Both updates are available in early access via the Matrox Imaging Library update service, with the official release scheduled for Q2 2020.
Share your vision-related news by contacting Dennis Scimeca, Associate Editor, Vision Systems Design
SUBSCRIBE TO OUR NEWSLETTERS
