Background: I am new to Tensorflow and AI and wish to try out Tensorflow in a Java based environment. Found Tensorflow has a Java API and tried it out but results don't seem to be similar to my Python implementations. I used Maven for my Java implementation.
I had retrained a dataset using Tensorflow in Python by following the Tensorflow for Poets tutorial.
In a Python based Tensorflow environment the dataset outputs values which are more or less correct. For example, if I have an image of a car and my classification options are car, bug,or building; it would output car.
But in my Java equivalent which I tried using the Tensorflow Java API and Image label demo; the datasets I produced earlier are readable but the outputs are completely different from the Python implementation. Unlike the Python example, if I throw a car it would output it as a bug or building.
I had tried the output data on several different machines with processors and gpus from different generations to see if it would affect the output; but it appears to be exclusive with the language/API.
Any ideas as to why the two implementations using the same dataset
produce completely different outputs?
And is there any way I could get my Java version to output similar
results to my Python example?
Copyright License：
Author:「nickzor」,Reproduced under the CC 4.0 BY-SA copyright license with link to original source & disclaimer.
Link to：https://stackoverflow.com/questions/53158340/tensorflow-dataset-produced-in-python-has-different-readings-in-tensorflow-java