README.md

Procedure

• Download data and unzip as it train, test, validation

• Name the csv's train.csv, test.csv, validation.csv

• Create directories train_resized, test_resized, validation_resized

  • Inside each of these make the directories pos, neg

• Complete folder structure:

  out/
  
  preview/
  
  train_resized/
  ├── neg/
  └── pos/
  
  test_resized/
  ├── neg/
  └── pos/
  
  validation_resized/
  ├── neg/
  └── pos/
  
  train/
  
  test/
  
  validation/
  

• call python scale.py

• call python split_data.py

• call python train_network.py noLoad if you don't want to load weights, other wise call python train_network.py path/to/weights.h5

• call python keras_to_tensorflow.py nameOfYourTrainedH5.h5 Note: Do not include the out/ just write the name of the file in the out folder

• call python path/to/strip_unused.py --input_graph=out/tensorflow.pb --output_graph=out/tensorflow-optimized.pb --input_node_names=input_1 --output_node_names=predictions/Softmax --input_binary=true