COMP502 – Foundation of IT Infrastructure
Semester 2, 2019
Assignment Part 3/3: Machine Learning and Deep Learning
Total Marks 100 (Contribution to the final mark: 20%)
Due: 23:59 pm, Sunday. 13 October 2019 (end of AUT calendar week 11).
Late penalty: Late submissions will be penalised at 5% of the actual assignment mark for every calendar day. E.g. assignments submitted 5 minutes late are penalised as one day late. Late work of 5 or more days will not be marked.
This assignment will help you understand the basic idea of Machine Learning and Deep Learning using TensorFlow and Google Colab. This assignment ask you to develop a simple Deep Learning Model to recognise one of the ten native birds of New Zealand. The ten birds are “Bellbird”, “Kahu”, “Kereru”, “Kiwi”, “Kokako”, “Piwakaka”, “Riroriro”, “Ruru”, “Tieke”, and “Tui”.
- The assignment will not be marked if:
- It contains any form of malware (e.g. computer virus)
- Submitted in non-requested compression format (e.g. 7zip, rar, …)
- Not submitted in a correct file format (see section: REPORT SUBMISSION INSTRUCTION)
- Keep a backup copy of your assignments to be:
- uploaded to “Turnitin” anti-plagiarism service – if requested.
- submitted as a hard copy – if requested.
Instruction to Start
- Download the attached file: Assignment_3_FTIT.ipynb
- Upload it to Google Colab: https://colab.research.google.com/
- Click File >> Upload Note Book
- Follow the instruction of the file; ask the TA in the lab for more details.
Instruction for Submission
The assignment must be submitted on AUT ONLINE / Blackboard in soft copy at:
Assessments > Current Assessments > Assignment Part 3/3: Machine Learning and Deep Learning.
- Rename your assignment to LAST NAME_Student-ID.ipynb. Then upload it to AUT ONLINE / Blackboard.
- Also, print the output to PDF file LAST NAME_Student-ID.pdf and upload to AUT ONLINE / Blackboard.
- Note that RAR, 7zip or virus contaminated files will not be marked (automatic failure)
Part 1: Test an ImageNet classifier using pre-built Models (30 marks)
Run it on three single images (10 marks each)
- Take three close-up pictures of your choice; they can be your pet, your vehicle, an animal you see in the zoo, your music instrument, your furniture, etc.
- Then, let’s make the computer recognise your pictures.
- Run the below codes to display the classification results. Are the results correct?
Part 2: Collect a Dataset to build a model to recognise one of the 10 NZ birds (50 marks)
Collect dataset; each correct image input is worth 0.5 marks.
- Start by downloading the following file: http://cerv.aut.ac.nz/cv/NZ_birds.zip
- Extract the file to a folder and copy it to your local drive or Google Drive.
- Enter your ID in the required field and run it, to find out which bird you are required to build recognising model, e.g. Kiwi.
- Download at least 100 pictures of the chosen bird and store them all in the correct folder inside NZ_birds folder.
- The images must be cropped just to contain the bird, and stored in JPG files.
- The size of each image should not be larger than 50 KB (resize it).
- After that, you can upload the entire folder NZ_birds to the home folder of Google Drive: https://drive.google.com/drive/u/0/my-drive (Click New >> Folder Upload).
- One mark for each correct picture.
- Create a shared link to your folder, right-click on NZ_birds folder and click Get a shareable link, and copy the link to the required field.
- Run the code, follow the instructions in the .ipynb
Part 3: Train your own model from the new data set in NZ_birds folder (20 marks)
Follow the instructions in the .ipynb file, and print out the results.
- Use compile to configure the training process.
- Plot the Loss and Accuracy of the training process.
- Show the prediction and display the results.
Comment on the prediction results: is it good or bad?
Part 4: Extend your model to build an object detection application on a Mobile phone (Bonus 25 marks – 5% towards your internal grades)
Extend the above parts to build a program to detect the selected bird within an image, with bounding boxes on your phone (iOS or Android). You can use any tools, any model architectures, and any platforms. One tool can be found here: https://www.tensorflow.org/lite/models/object_detection/overview
To claim these 25 marks (or 5% towards your internal grades), show the working programs, codes, videos, or screenshots, to your lecturers (email [email protected] for appointment). This part is specifically offered to A+ students only. No further instruction is given.
本网站支持淘宝 支付宝 微信支付 paypal等等交易。如果不放心可以用淘宝交易！
E-mail: [email protected] 微信:dmxyzl003