tensorflow-data

An small example how to use tensorflow data (tf.data)

Usage

In order to try this repository you clone it on your drive. You'll probably need 12-20Gb of disk because of the large amount of image data.

First of download the image Dataset and put the picture in a structure where you have

./PetImages/Cat/*.jpg
./PetImages/Dog/*.jpg

I've not pin-point everything required to run this test scripts because I had most of it installed already. But you need to install tensorflow, opencv2 and numpy atleast.

pip install tensorflow opencv2-python numpy

Then you run the create_dataset.py in order to create the train, test and validation tfrecords.

python create_dataset.py

Lastly you can train your model using the training script.

python train.py

If you have any question or suggestion the just reach out. Open an issue and I'll look into it.

Tensorflow lite model.

In order to use this model on a tensorflow lite enabled device you need to freeze your model using this command

freeze_graph \
  --input_graph=./model2/graph.pbtxt \
  --input_checkpoint=./model2/model.ckpt-81852 \
  --input_binary=false \
  --output_graph=/tmp/frozen.pb \
  --output_node_names=input_tensor,output_pred

In order to convert your model you need a tool called toco (Tensorflow Lite Optimizing Converter). Use the command below to build this tool in your tensorflow directory.

bazel build //tensorflow/contrib/lite/toco:toco

After that you convert it into a tensorflow lite model using the command below inside of your tensorflow directory.

 ./bazel-bin/tensorflow/contrib/lite/toco/toco 
   --input_file=/tmp/frozen.pb 
   --input_format=TENSORFLOW_GRAPHDEF 
   --output_format=TFLITE 
   --output_file=/tmp/cat_vs_dogs.tflite 
   --input_arrays=input_tensor 
   --output_arrays=output_pred 
   --input_shapes=1,224,224,3

Modifying the demo provided by google you can then test your inference on your device.