TensorFlow Tutorial - Transfer Learning


Notes by Magnus Erik Hvass Pedersen: https://github.com/Hvass-Labs/TensorFlow-Tutorials/blob/master/08_Transf...

We saw in the previous Tutorial #07 how to use the pre-trained Inception model for classifying images. Unfortunately the Inception model seemed unable to classify images of people. The reason was the data-set used for training the Inception model, which had some confusing text-labels for classes.

The Inception model is actually quite capable of extracting useful information from an image. So we can instead train the Inception model using another data-set. But it takes several weeks using a very powerful and expensive computer to fully train the Inception model on a new data-set.

We can instead re-use the pre-trained Inception model and merely replace the layer that does the final classification. This is called Transfer Learning.

This tutorial builds on the previous tutorials so you should be familiar with Tutorial #07 on the Inception model, as well as earlier tutorials on how to build and train Neural Networks in TensorFlow. A part of the source-code for this tutorial is located in the inception.py file.

Resource Type: