Description
In this project you are asked to do tranfer learning on the “flowers” dataset from Kaggle. The
dataset has 13 classes. We will assume that classes are described in alphabetical order:
classes = {astilbe, bellflower, black-eyed susan, calendula, california poppy,
carnation, common daisy, coreopsis, dandelion, iris, rose,
sunflower, tulip}.
You may find the example code in the following link to be useful:
https://www.tensorflow.org/tutorials/images/transfer learning
In designing your program you may use all of the functionality of OpenCV, and deep learning under
the following conditions:
• You must use tensorflow.
• You must not use pytorch for this project.
• You must use the backbone models from:
https://www.tensorflow.org/api docs/python/tf/keras/applications
• You cannot use any special libraries or software packages. Please contact the TA if you are
not sure if something should be considered “special”.
Provided files
• flowers.zip file for training data.
• The python script proj2 test.py is provided as basic for testing program. Test program
read in test folder and csv file for their labels.
• Sample test folder as flowers test.zip and the corresponding csv file flowers test.csv files
are provided for testing.
Evaluation
• We will use test accuracy as the measure for evaluation.
• We will create a test folder containing test images and their corresponding labels. The images
will not be the same as those in the provided flowers folder, but similarly collected from
the web.
1
What you need to submit
1. Python or ipython notebook source code of your program. Please name it proj2.py.
2. Documentation that explains your approach.
3. Your model, weight files, your testing code and instruction on how to set up environment for
testing your code. Please put in any pre-processing you do in your test code. Please name
this as proj2 test.py.
These items should be put in a zip file named with your netid and submitted on elearning. For
example, if your netid is xyz1234 you should submit a zip file named xyz1234.zip.
2 CS6384 Transfer Learning
