Lab Project 10: Practice with Transfer Learning: Modifying and training Googlenet to classify merchandise

Use the deepNetworkDesigner to retrain googlenet to recognize and classify the five merchandise classes shown above (cap, cube, screwdriver, playing cards, and torch).

The data (pictures) used for training are located here:
http://socsci.uci.edu/~saberi/psych149/images/

Download the folder called MerchData.zip to the folder containing the program you are writing for this project. Open the zip file, and copy and paste the folder MerchData to the same folder.  Your program will access the images in this folder. Take a look inside the folder and you'll see 5 subfolders, each containing 15 images of each class (for a total of 75 images).  These images are provided by Matlab.   The labels for training (i.e., correct answers) are just the names of the folders, so the label or correct answer for the Cubes is the folder name "MathWorks Cube" (you can change the folder name to just "Cube" if you'd like, but it's not necessary).

At the end of the program,  randomly pick four of the images and see if the trained network correctly classifies them.

Step 1

Step 3

Step 6