Using Google Cloud AutoML Edge Image Classification Models in Java

Showcase

I built a tensorflow model with Google AutoMl Vision , downloaded it for local use und wrote a java showcase to execute it:

The showcase processes all images (jpg,png,gif) in the given directory (--scanDir) - the used model is saved in the jar and will be copied to the given directory (--temp).

Usage (if you can't build it by yourself, take the showcase release) :

java -jar java_tf_image_classification-0.0.1-SNAPSHOT-all.jar --temp /tmp/model --scanDir /dirContainingPics

jar is compatible to run on Mac, Windows and Linux

Hints:

The model has been trained to classify following classes: motorbike, airplane, dog The model is only a showcase and is not intended to be a good model.

Example output:

OK   tenor.gif -> {motorbike=0.03536664, airplane=0.03536664, dog=0.92926675} -> dog
OK   dog_PNG50388.png -> {motorbike=0.050735015, airplane=0.1696087, dog=0.7796563} -> dog
OK   Ì.jpg -> {motorbike=0.034807786, airplane=0.06581213, dog=0.8993801} -> dog

Google AutoMl Vision

Model quality

Model export ["Export your model as a TF Saved Model to run on a Docker container." or with this showcase in java ;-)]

my thanks go to

• https://heartbeat.fritz.ai/using-google-cloud-automl-edge-image-classification-models-in-python-92f2885c767
• https://github.com/tensorflow/java