andrea-zanella / keras-transfer-learning-inception-resnetv2

Keras and tensorflow transfer learning, starting from the pre-trained inception_resnetV2 model

keras-transfer-learning-inception-resnetv2

Requirements

You will need 2 files from the official keras repo: https://github.com/fchollet/keras :

keras/applications/inception_resnet_v2.py keras/applications/imagenet_utils.py Place(replace) them in your keras/applications folder (usually at "/Library/Python/2.7/site-packages/keras" but it could be different.

You have to replace these two files so you can easy use the inception resnet v2 model.

You will also need the dataset, u can find the one I used for this example at https://mega.nz/#!Bo42wLKK!yyR7gXJSFOSwy-W7R-TVbSfBmict5TdUvTuiSvEOZSE

As you can see the dataset follow this structure: Dataset
├── cats
├── dogs
├── horses
└── humans

You can use the sample dataset from the link or you can also change it but u have to follow this structure.

Edit: Looks like the dataset is gone, you can use the flower dataset in order to test this code. Just open a new bash terminal (if you are on windows open Git bash) and type:

curl -LO http://download.tensorflow.org/example_images/flower_photos.tgz
tar xzf flower_photos.tgz

This will download and extract the dataset in the folder.

Quick start

Run evaluate.py If it's the first time it will train the last layer of the model with the dataset you choose. Otherwise it will just load the model previously saved.