Google Code-in

In this task, you will do image classification using a custom model and transfer learning.

Task Description:

1) Read about transfer learning using TensorFlow.

2) Pick a data set for image classification which is not trivial. The cats and dogs data set, for example, would not be acceptable for this use case.

3) Make a model for Image Classification using transfer learning, trying at least 2 pre-trained models, and learn why one works better than the other for that particular data set.

4) Make a custom model using TensorFlow 2.0 and compare it with the model trained in the previous step.

5) Make utility functions looking at various comparisons and various outputs of a model in a visual manner using matplotlib/seaborn such as :

• Comparing Training time, loss curves of 2 models
• Images that are predicted incorrectly by a model and with what accuracy.
• Running prediction on a folder.
• Running prediction on an image.
• Calculating metrics such as precision, recall and accuracy for the testing set.