perClass featured as one of the most comprehensive machine learning tools in Vision Systems article
The last issue of the Vision Systems Design journal brings an article on advanced machine learning technology for industrial vision applications. The article features PR Sys Design and perClass among the major players in the machine vision field.
It also highlights in more details key points of our design methodology, e.g. use of multiple classifiers to develop accurate systems, understanding the problem by building classifiers and fundamental importance of performance optimization to find “good-enough” solutions.
Spring 2015 course scheduled for the week 13-17 April
The spring perClass course will take place in the week 16 (13-17 April 2015) in Delft, The Netherlands.
4.4 release with one-class SVM and significantly faster multi-class ROC
Happy to announce perClass 4.4 that brings several improvements such as support for one-class SVM classifiers and significantly faster multi-class ROC estimation. Number of improvements address the new graphical foundation of Matlab 2014b. This release also introduces shape-based object-level feature extraction from images and super-easy back-projection of object labels to original images. Check-out the full list in the Release notes and Enjoy!
updated 4.3 version available with multi-class SVM fix
We have issued a fix for sdsvc as a part of 4.3 24-jun-2014 release. It is resolving an issue with incorrect offset of RBF and polynomial SVM that impacted mostly multi-class SVM classifiers. Two-class SVMs followed by ROC to choose an operating point were not affected.
perClass demo of defect detection at Vision and Robotics fair
Pictures from Vision and Robotics fair 2014 are in the blog
Autumn course scheduled on 24-28 November 2014
The Autumn perClass course will take place in Delft, The Netherlands in the week 24-28 November 2014. Details and registration
4.3 release with Matlab documentation and imaging improvements
Happy to announce perClass 4.3 release. It provides all documentation in Matlab format, new logistic classifier and many improvements of the imaging framework.
Image feature extraction is greatly simplified, there is now a faster connected component analysis implementation and a new object feature extraction. Typical image recognition tasks such as object labeling, feature extraction and object classification are now significantly faster and easier to use.
In addition to new features, 4.3 comes with many usability improvements and fixes including some hard-to-find issues (sdimage label layer problem in Windows, figure positioning bug in Matlab 2013 and higher on LInux, precision timers on MS Windows and more). See Release notes for the complete list.
April perClass course for R&D specialists in Delft
In-house training course in Wageningen
In the beginning of April, we organized training course in-house in the Wageningen UR plant sciences group. More details in the blog
4.2 release brings improved interactive scatter plot
perClass 4.2 release brings number of usability improvements. Working with subsets of samples in a scatter plot is now simpler and more visual. It is now possible to invert and remove individual filters, tag samples and perform many common tasks entirely from the scatter plot (e.g. creating new label sets from another set and labeling groups of samples). The confusion matrix can be now rendered in a figure and understood in a glance. More details in the release notes. Enjoy!