Skip to content

Latest commit

 

History

History
59 lines (36 loc) · 3.07 KB

Challenge04.md

File metadata and controls

59 lines (36 loc) · 3.07 KB

Challenge 4: Following the Marked Trail

Background

Previously, you build a convolutional neural network (CNN) to classify product images. A CNN typically consists of multiple convolutional, pooling, and drop layers that extract features from images, and one or more dense fully-connected layers that map those features to classes.

Transfer Learning is a commonly used machine learning technique in which you can leverage the feature extraction layers from an existing model, and add your own fully-connected layer to predict classes from the extracted features.

Prerequisites

  • A Data Science Virtual Machine (DSVM)
  • The resized gear image data fom the previous challenges.
  • An installation of the latest version of your chosen deep learning framework(s) based on the References section below.

Challenge

There are three elements to this challenge:

  1. Explore a sample transfer learning model.
  2. Use transfer learning to train a classifier based on an existing model.
  3. Use your model with new data.

1. Explore a sample transfer learning model

Explore the notes and code in the 04-Transfer Learning (framework).ipynb notebook in the ready2019/notebooks folder to see an example of using transfer learning to train a CNN.

2. Use transfer learning to train a model

Create a new CNN by using transfer learning to build a classifier on top of the feature extraction layers defined in an existing model.

Hints

    • Use the Python 3.5 kernel in Jupyterhub on your DSVM.
  • Base your initial solution on the code in the sample notebook.
  • You can use any base model supported by your chosen deep learning framework.
  • You may need to resize the images to match the size used to train the base model you select.

3. Use your model with new data

Use your model to predict the class of at least five images that are not included in the gear dataset. You can use the same five images you found in the previous challenge.

Success Criteria

  • Successfully train a CNN based on an existing trained model.

  • Show predictions for the five images you identified in the Challenge section, like this:

    Gear predictions

    (Note: Your model is not required to predict the correct class for all of the images, but it would be good if it does!)

    When your coach has verified your team's solution, you have completed all of the challenges - congratulations!

References