Tensorflow: Use Retrained categories in Android Camera Demo

[hongyehenry]

GitHub issues are for bugs / installation problems / feature requests.
For general support from the community, see StackOverflow.
To make bugs and feature requests more easy to find and organize, we close issues that are deemed
out of scope for GitHub Issues and point people to StackOverflow.

For bugs or installation issues, please provide the following information.
The more information you provide, the more easily we will be able to offer
help and advice.

Environment info

Operating System: Docker on Windows

Installed version of CUDA and cuDNN:
(please attach the output of ls -l /path/to/cuda/lib/libcud*):

If installed from binary pip package, provide:

1. Which pip package you installed.
2. The output from python -c "import tensorflow; print(tensorflow.__version__)".

If installed from source, provide

1. The commit hash (git rev-parse HEAD)
2. The output of bazel version

Steps to reproduce

1. Successfully retrain new data with Docker. Get the two retrained files retrained_graph.pb and retrained_labels.txt
   https://codelabs.developers.google.com/codelabs/tensorflow-for-poets/index.html?index=..%2F..%2Findex#5
2. Try to install Bazel on windows but failed. But I found an Android Demo which doesn't need Bazel. This one works well on Windows. It can recognize images into 1000 classes provided by ImageNet in an Android real-time camera.
   https://github.com/miyosuda/TensorFlowAndroidDemo
   

   What have you tried?

Reference； https://github.com/tensorflow/tensorflow/issues/1269
Based on the suggestions, I tried the following things:
1.Find the coded_stream.h in the google/protobuf section of your tensorflow build and modify the 64 to 256 in the following line:
static const int kDefaultTotalBytesLimit = 64 << 20; // Change the 64 to 256 MB
2.Modify only the input_size to 299 and the image_mean to 128 in the TensorflowImageListener.java
3.Go to tensorflow_jni.cc in the android demo and modify as follows:

  input_tensor_mapped(0, i, j, 0) =
      (static_cast<float>(src->red) - g_image_mean)/g_image_mean;
  input_tensor_mapped(0, i, j, 1) =
      (static_cast<float>(src->green) - g_image_mean)/g_image_mean;
  input_tensor_mapped(0, i, j, 2) =
      (static_cast<float>(src->blue) - g_image_mean)/g_image_mean;
  ++src;

std::vector > input_tensors(
{{"Mul", input_tensor}});

std::vectorstd::string output_names({"softmax"});
4.Do the following changes in TensorflowImageListerner.java:
private static final String MODEL_FILE = "file:///android_asset/retrained_graph.pb";
private static final String LABEL_FILE =
"file:///android_asset/retrained_labels.txt";

1. Build the android demo

Logs or other output that would be helpful

(If logs are large, please upload as attachment).
08-01 17:36:50.015 14978-15121/org.tensorflow.tensorflowdemo A/native: jni_utils.cc:107 Check failed: message->ParseFromZeroCopyStream(&adaptor)
08-01 17:36:50.015 14978-15121/org.tensorflow.tensorflowdemo A/libc: Fatal signal 6 (SIGABRT), code -6 in tid 15121 (ImageListener)

I know this is a problem due to the incompatible variables of Inception v3 and Inception 5h. The Android Demo model I use is Inception 5h, But the model I used to train my new data is Inception 3v. I've tried to edit the variables mentioned here: https://github.com/tensorflow/tensorflow/issues/1269.
But I still get the same errors. Anyone could help me explain how to adapt my new trained data running in the Android Demo?

[poxvoculi]

This sort of question is better asked on StackOverflow. This forum is for bug reports and similar. Please ask your question there, tagging with 'tensorflow'.

If you're having trouble with bazel, you might try tensorflow makefile.