Exporting label images provides the label masks as PNG files including class names meta-data. This is useful when using perClass Mira as a precise annotation tool for an external machine learning training work-flow.



A destination directory can be selected and an optional scan name suffix can be specified. This is useful to distinguish different groups of scans, for example, due to their training/test status or product variant. We may view the exported images conveniently in the Windows Explorer:


Class name meta-data

The exported images contain class name meta-data.

Accessing meta-data using TweakPNG

The free TweakPNG utility can be used to view PNG file meta-data: http://entropymine.com/jason/tweakpng


Accessing meta-data in Matlab

Using the imginfo command in Matlab, we can access the meta-data information stored by perClass Mira. We need to extract the OtherText property:


>> s=imginfo('natural_objects1.png');

>> s.OtherText


ans =


  3×2 cell array


    'ClassCount'     '6'                                         

    'ClassFormat'    'perClass Mira 1'                           

    'ClassNames'     'Unknown,background,leaves,nuts,shells,wood'