Matrox Imaging updates SDK to expand deep learning and 3D
Additions and enhancements to the Matrox Imaging Library (MIL) X expand deep learning and 3D capabilities, says Matrox Imaging.
For classification and 3D tools, the Service Pack 5 (SP5) has new hand-eye calibration for robot guidance, together with improvements to the SureDotOCR tool. There is also a revision of the CPython interface.
MIL X uses deep learning technology, specifically the convolutional neural network (CNN) and variants, to perform image and pixel classification for identification and defect detection tasks, respectively.
SP5 allows users to benefit from a simplified workflow for training a deep neural network, with corresponding new models producing finer results in the case of segmentation. There is also a new inference engine enabling higher performance and broader hardware support. MIL X provides the infrastructure and interactive environment for building the training dataset and monitoring and analysing the training process, explains Matrox. SP5 streamlines training by automatically sub-dividing the training dataset, offering pre-set augmentations and proposing a neural network model to start with.
Matrox Imaging specialists are available for hire to train a deep neural network on a user’s behalf using the infrastructure and interactive environment.
There is also a set of tools for performing 3D processing and analysis. Enhancements in SP5 include filters for noise removal. There are also two new tools, one for shape-finding and one for blob analysis, which expand a user’s ability to locate and measure features and objects in a point cloud.
Another tool, 3D calibration, facilitates the use of a camera or sensor within a robotic cell. Users can opt for eye-to-hand or eye-in-hand calibration, both of which accurately relate what a camera or sensor sees, i.e., the object pose established using other MIL X tools, to a robot controller for guidance.
MIL X SP5 also includes improvements to Matrox SureDotOCR tool for reading dot matrix text produced by inkjet printers and dot peen markers. The updated tool now supports text of variable length as well as variation in dot size and overlap.
A reworking of the CPython interface features NumPy support to improve the experience and efficiency, says Matrox Imaging.
Summing up the enhancements, Pierantonio Boriero, director of product management, at Matrox Imaging commented: “New neural network models, inference engine, and simplified deep learning training, plus improved 3D tools all combine to present the most capable version of this SDK yet offered.”
SP5 for MIL X is now available.
