COMP90086 Computer Vision Final Project

👥 Team Members

• 📌 Lujie Ma
• 📌 Tao Yu

📖 Project Overview

Objective

Our primary goal is to develop an efficient algorithm for the "Totally-Looks-Like" challenge, aiming to match similar looking images from a dataset.

Description

We've embarked on this journey to craft an algorithm that can adeptly match an image from a "Totally-Looks-Like" pair with its counterpart from a myriad of potential matches. Our journey is deeply rooted in critical analysis and rigorous evaluation using the dataset provided. The outcomes will be gauged through error analysis, unveiling the robustness and fragility of our approach.

Dataset

Fetch the dataset from the link below: 🔗 Dataset Download

🔍 Instructions:

1. Grab the dataset using the link above.
2. Decompress the fetched file.
3. Relocate the extracted content to this project's root directory.

📂 Code Organization

• final_code.ipynb: Core logic and operations reside within this Jupyter notebook.
• Siamese.ipynb: The blueprint for our Siamese Network can be found here.
• components: This repository shelters our project's comprehensive report and vital utilities:
  • DataProcessor.py: Entrusted with data preprocessing.
  • FeatureExtractor.py: An encyclopedia for feature extraction.
  • ModelParametersCounter.py: The math wizard counting model parameters.
  • SimilarityEvaluator.py: A judge evaluating image resemblances.
  • VectorComparator.py: The expert distinguishing vector similarities.

🚀 Running the Code

Ensure the following Python packages are installed:

pip install -r requirements.txt

Subsequently, you can execute the Jupyter notebooks, which are pre-configured to run by default.

📃 Report

View the report here: Report.pdf