0

I have around 100,000 images that represent Music Artwork and I am trying to modify the Android Tensorflow example following

https://jalammar.github.io/Supercharging-android-apps-using-tensorflow/

and the project

https://github.com/miyosuda/TensorFlowAndroidDemo

I think I need to create a new TensorFlow graph like tensorflow_inception_graph.pb , but I am wondering if there is any easy way to train and create a graph like that, just like drag and drop the images and get the file after few hours and then just integrate?

Can someone describe a bit the steps needed to achieve this task?

Kenenisa Bekele
  • 835
  • 13
  • 34
  • Perhaps start with https://github.com/tensorflow/models/tree/master/inception followed by http://stackoverflow.com/questions/34343259/is-there-an-example-on-how-to-generate-protobuf-files-holding-trained-tensorflow – Morrison Chang May 10 '16 at 16:22

1 Answers1

0

What is your goal? Are you trying to identify objects? Forget for a moment that you can use TensorFlow, what do you want as a result of the analysis?

After defining your goal try to perform a research to find papers that are solving similar problems, and after what model they are using try to implement your own model using similar architectures for your neural net. Try to understand the design decisions behind the different approaches.

You need to train your own model using tensorflow on a PC or on a cluster of servers, or a cloud service. You can't use Android for that now a days (for training the model), after train the model you can use the model (the graph) in an Android App, in the same way that in the official example.

Matias Molinas
  • 2,246
  • 15
  • 11