|
Image classification technologies are constantly gaining importance in industrial and personal use,
similar to mobile devices and applications. In parallel, the authors observe the popularity of
developer tools with lower entry levels than classical programming languages, like Java, the tools
being a part of the so-called “Do It Yourself” approach. One such tool is App Inventor with recently
added support for image classification. Due to the abovementioned factors, the authors have decided
to check App Inventor performance against Java for Android in image classification based on the
TensorFlow library. The purpose of such a test was to determine whether the App Inventor is able
to classify images in the mobile environment of limited resources with reasonable outcomes.
Analogous applications were built in each technology, then the applications were run on the same
set of images depicting a few different kinds of objects. The effectiveness and time of image
classification were measured. It was revealed that both applications had comparable effectiveness
of image classification and most of the objects were correctly recognised. Unfortunately, the
application in App Inventor was on average two times slower than the other application, although
the time of a single classification did not exceed 0.3s. It could be concluded that when deciding
about App Inventor usage, it has to be considered what is more important – the performance whether
the simplicity of coding applications.
|
