This release includes all features which were previously only available in the Early Adopter versions. Deep Learning Tool 0.5 will include the following features and improvements:
Training for classification – Set all important parameters and execute a training based on your labeled data – directly within the Deep Learning Tool.
Evaluation – Evaluate and compare your trained networks directly in the tool with heatmaps and confusion matrices and generate HTML reports.
Data split handling – When configuring a training for classification you can assign images to custom splits between the training, evaluation, and testing datasets. Class weights can now be manually adjusted.
Filtering of project images – Filters apply to the Gallery, Image, and Review pages, as well as the HDICT export and the statistics. You can use several predefined quick filters and create your own filters. Currently, you can filter images by name, file path, and by label status and used label classes.
How to get it?
Users who already installed version 0.3.1 can simply update to the latest version via the MVTec Software Manager (SOM). Users of DLT 0.4.3 EA can install version 0.5 via SOM as well, but must select the corresponding product line from the "available" tab.
In both cases, be sure to first hit “refresh from server” to get the latest list of available versions.
Users of older versions must first download the new MVTec Software Manager from the MVTec website to install this new release, since the old MVTec Package Loader and Manager has been discontinued.