Tensorflow Lite: No module named tf.contrib.lite

Question

I tried to convert my already existing frozen graph, which is saved in a .pb-file, with the following code (tf_lite_converter.py):

#!/usr/bin/env python
import sys
import tensorflow as tf

from tf.contrib.lite import convert_savedmodel

convert_savedmodel.convert(
    saved_model_dir="/frozen_inference_graph.pb",
    output_tflite="/TF_Lite_Model")

When running the code with

python tf_lite_converter.py 

in my anaconda environment, it gives me the error:

ImportError: No module named tf.contrib.lite

My goal is to get a tensorflowlite-model out of my .pb-graph to use it in an Android application. I already tried to build tflite with toco through bazel, but maybe (or most likely) I did something wrong or it was not the right way to fix the problem.

Referencing Video from Tensorflow-Developers: https://youtu.be/FAMfy7izB6A?t=11m49s

Answer 1

The code in that video is probably from an internal development version.

convert_savedmodel has been renamed to convert_saved_model in https://github.com/tensorflow/tensorflow/commit/db076ca01f12368c9476fa4db9d87756f22f9670

The following seems to work for tensorflow 1.8:

from  tensorflow.contrib.lite.python import convert_saved_model

convert_saved_model.convert(saved_model_dir="/frozen_inference_graph.pb",output_tflite="/TF_Lite_Model")

The following is for tensorflow built from the current master (method and parameter have been renamed):

from  tensorflow.contrib.lite.python import convert_saved_model

convert_saved_model.tflite_from_saved_model(saved_model_dir="/frozen_inference_graph.pb",output_file="/TF_Lite_Model")

Answer 2

None of the above things worked for me.

I downgraded to Tensorflow 1.7 and converted the .pb model to .tflite model using toco.

$ pip install --upgrade "tensorflow==1.7.*"

$ toco \
  --input_file=tf_files/retrained_graph.pb \
  --output_file=tf_files/optimized_graph.lite \
  --input_format=TENSORFLOW_GRAPHDEF \
  --output_format=TFLITE \
  --input_shape=1,${IMAGE_SIZE},${IMAGE_SIZE},3 \
  --input_array=input \
  --output_array=final_result \
  --inference_type=FLOAT \
  --input_data_type=FLOAT

Ref : https://codelabs.developers.google.com/codelabs/tensorflow-for-poets-2-tflite/#2 https://github.com/googlecodelabs/tensorflow-for-poets-2/issues/52 https://medium.com/@rdeep/tensorflow-lite-tutorial-easy-implementation-in-android-145443ec3775