ELEC4630 Image Processing and Computer Vision
Assignment 3
(Due date: Friday 24/5/2024 at 4pm)
Assignment report should include coding, results, images, and a verbal description of how you approached the problem. Some similar solutions can be found on the internet, but you won’t learn anything by copying these verbatim, and you may be flagged for integrity issues — none of us want this. However, in this assignment you are encouraged to find internet solutions and modify the code significantly to match your problem. This is fine and ethical as long as you cite your sources appropriately. You may also use generative AI technologies such as ChatGPT. In this case, you must declare what you used and how you used them. Also, the tutors will help you with the coding. Most importantly, have fun.
1. For the first competition we are going to look at
https://www.kaggle.com/competitions/paddy-disease-classification
Problem Statement
Rice (Oryza sativa) is one of the staple foods worldwide. Paddy, the raw grain before removal of husk, is cultivated in tropical climates, mainly in Asian countries. Paddy cultivation requires consistent supervision because several diseases and pests might affect the paddy crops, leading to up to 70% yield loss. Expert supervision is usually necessary to mitigate these diseases and prevent crop loss. With the limited availability of crop protection experts, manual disease diagnosis is tedious and expensive. Thus, it is increasingly important to automate the disease identification process by leveraging computer vision-based techniques that achieved promising results in various domains.
Objective
The main objective of this competition is to develop a machine or deep learning- based model to classify the given paddy leaf images accurately. We provide a training dataset of 10,407 (75%) labeled images across ten classes (nine disease categories and normal leaf). Moreover, we also provide additional metadata for each image, such as the paddy variety and age. Your task is to classify each paddy image in the given test dataset of 3,469 (25%) images into one of the nine disease categories or a normal leaf.
Hint:- look at Jeremy Howards solution here:
https://www.kaggle.com/code/jhoward/first-steps-road-to-the-top-part-1
and notebook 8 of course 22. Iterate to find the best solution. (15 marks)
2. The quality of AI-generated images has rapidly increased, leading to concerns of authenticity and trustworthiness.
CIFAKE is a dataset that contains 60,000 synthetically generated images and 60,000 real images (collected from CIFAR-10). Can computer vision techniques be used to detect when an image is real or has been generated by AI?
Design a deep learning model to classify images into either fake or real. Evaluate the performance of your model and analyze the impact of varying hyperparameters and architecture choices on model accuracy and training time. This dataset was released in 2023, so we really don’t know how well we can perform on this task. See if you can win Kaggle for UQ!
https://www.kaggle.com/datasets/birdy654/cifake-real-and-ai-generated- synthetic-images (15 marks)
(Total 30 Marks)
Assignment 3 Marking Scheme and Criteria
Q1
. Go through the lab sessions independently. Analyse the techniques for obtaining
good solution. Try some additional ideas in terms of data augmentation and learning
rates. See if you can do better than Jeremy,
o (15 marks)
Q2
. Classify real vs fake images. Appropriate discussion and approach.
Implementation and evaluation of several methods and architectures. Notebook coding quality, ease of reading text.
o (10 marks)
. Explanation justifying different architecture choices, Evaluation of the work of others. Recent literature search analysis. List of references. Quality of citations.
o (5 marks)