Deep Learning - Exercise 10

Learning goal: Learn how to use a pre-trained CNN

We have seen in the last exercise that training a deep CNN from scratch needs a lot of time. For this, it is not a bad idea to try to find a pre-trained CNN model and to try to make use of it.

1. Try to find a pre-trained CNN model

Try to find a CNN implementation and pre-trained model data in the WWW in some programming language (Java, C++, Python, ...) and within some Deep Learning framework.

2. Test the model on new data

If you have found a pre-trained CNN model and managed it to get it running on your system, test the model on new data it has (probably) not seen during its training phase to assess the performance of the pre-trained model and what you can expect when using this model as it is.

Central question: Can it recognize all the classes you are interested in?

3. Transfer-Learning

Try to continue training the model with new data!