VGG16-Model-Transfer-Learning

Transfer learning using tensorflow VGG16 image classification model

We have used the transfer learning VGG16 classification model to classify our own GOT dataset. After creating trasnfer codes using the existing model, we have added fully connected and activation layers to perform a higher layer classfication on our GOT dataset.

The model reaches validation accuracy of ~95% in 20 Epochs and is extremely fast!

It shows how transfer learning can be used to solve classification tasks, by building sleek and prodigiously fast models.

Steps:

Run the model using file "transferModel.py"

Overview:

• transferModel file imports createTransferCodes file
• createTransferCodes does following things:
  • download the VGG16 model data (.npy file)
  • apply transformations if required
  • create transfer codes and store them in the project directory

Note: Images transformations are created if number of images for any class are less than 500

References:

Image transformation: https://github.com/vxy10/ImageAugmentation
VGG16 Model: https://github.com/machrisaa/tensorflow-vgg
Progress bar: https://blog.shichao.io/2012/10/04/progress_speed_indicator_for_urlretrieve_in_python.html