README.md

inception-v3 image classifier

Simple parametized python script to use a fine trained Inception V3 model to classify images.

Based on:

• Tensorflow example https://www.tensorflow.org/versions/r0.11/how_tos/image_retraining/index.html#training-on-your-own-categories
• This great article on codeLabs https://codelabs.developers.google.com/codelabs/tensorflow-for-poets/#0

NOTE: This version will work only with TensorFlow-0.9.0-devel!

Dependencies

• Python >= 2.7
• Tensorflow 0.9.0-devel (see https://www.tensorflow.org/versions/r0.9/get_started/os_setup.html)

Usage:

• Fine train the Inception v3 model using the train.sh script:

  $ ./train.sh --tf_bin=/path/to/tensorflow/installation --tf_data=/path/to/images/data/folder.

  You can put any number of sub-directories in your data folder, inception will be fine trained to classify any images in categories define by those sub-subdirectories.

  i.e.
  /cat
   -> Persian
   -> Bengal
   -> Burmese
   -> Ragdoll
  

  Will train inception to classify any picture into those 4 Cat's breeeds cathegories.

• Classify your images with label_image.py:

  $ ./label_image.py --datafolder=/tensorflow --image_path=img/cat.jpg
  Persian (score = 0.88331)
  Bengal (score = 0.11669)
  Burmese (score = 0.23879)
  Ragdoll (score = 0.17469)