Tensorflow Java API - A Complex Example

Question

I already asked the question as a github question, but was redirected here. I saw an example to import a model created and trained in Python, imported into Java code and used for predictions. However, I had problems understanding what was actually going on, especially in this block and the declaration of the GraphBuilder class between lines 156-207 . Can someone please give them some explanation?

Also, I know that the Java API is still under development. However, I would be interested if more complex examples can be seen, if possible, including:

• import the model into Java and then do the model training

• implementing, training, evaluating, saving, loading a model from scratch in Java using Tensorflow

Does anyone have such an example and are willing to share it?

Thanks for the help!

Greetings

Peter

Answer 1

The specified code block generates a TensorFlow graph to “normalize” the image so that the image can be fed to another TensorFlow graph (start). Python achieves the equivalent of something like this:

image = tf.cast(tf.image.decode_jpeg(input, channels = 3), tf.float32)
batch = tf.expand_dims(image, 0);
resized = tf.image.resize_bilinear(dims_expander, [input_height, input_width])
normalized = tf.divide(tf.subtract(resized, [input_mean]), [input_std])

Python TensorFlow (, tf.cast, tf.image.decode_jpeg ..) ops TensorFlow. Java API, , GraphBuilder.

, .

, , .

Answer 2

, TensorFlow Java API . YOLOv2. : https://github.com/szaza/tensorflow-example-java.

- Spring Gradle, . : https://github.com/szaza/tensorflow-java-examples-spring.

Google Cloud, : http://35.229.93.105:8080/ - Heroku. yolo. . : https://still-crag-64816.herokuapp.com/