Siamese Network s43032053

[richardchantra]

Dear COMP3710 teaching team,

Name: Richard Chantra
Student Number: s43032053
Project: #6 Siamese Network

I have 5 files for this project:
modules.py: Defines Siamese network, MLP classifier, loss functions, and evaluation
dataset.py: Manages data loading, augmentation, and data balancing
train.py: Trains the Siamese Network and MLP classifier
README.md: Project documentation

Thank you

[hanemma7moud]

This is an initial inspection, no action is required at this point

Recognition Problem : total : 17

1. Solves problem: The solution is appropriate for the problem but suffers from poor performance due to class imbalance and possibly an inadequate feature-extraction backbone.
   The resulting manifold is not well-separated.
   The classifier performs well in terms of overall accuracy, but:
   Sensitivity: Low at 32%, indicating poor performance in identifying malignant cases. (3)
   Recall: Excellent for benign cases at 99%, but only 1% for malignant cases.
   Different rebalancing strategies have been adopted, but further experimentation is needed to achieve balanced performance, particularly to improve the model's ability to classify malignant cases.

2. Implementation functions as intended but A pre-trained ResNet is used for feature extraction (2)

3. Good design: Well-designed (1)

4. Commenting: Clear and sufficient comments throughout the code. (1)

5. Difficulty: Hard (10)

Note:

• A significant factor affecting the separation between classes might be the feature-extraction backbone’s strength and the chosen data balancing method.
• Lack of Data Diversity: Resampling merely duplicates existing samples without adding new information, increasing the risk of overfitting, especially with small, diverse minority classes. In medical imaging, data diversity is essential for the model to learn robust, generalizable features.

[gayanku]

Marking

Good/OK/Fair Practice (Design/Commenting, TF/Torch Usage)
	Good design and implementation.
	Spacing and comments.
	Header blocks.
Recognition Problem
	OK solution to problem. Model affected by imbalance.	-3
	Driver Script present.
	File structure present.
	Good Usage & Demo & Visualisation & Data usage.
	Module present.
	Commenting present.
	No Data leakage found. Has data splitting.
	Difficulty : Hard. Siamese network (Hard Difficulty)
Commit Log
	Good Meaningful commit messages.
	Good Progressive commits.
Documentation
	Readme :Good.
	Model/technical explanation :Good.
	Description and Comments :Acceptable.	-1
	Markdown used and PDF submitted. PDF Checked.
Pull Request
	Successful Pull Request (Working Algorithm Delivered on Time in Correct Branch). On time.
	Feedback action required: Feedback marks possible +2 if the requested changes are made. Merge conflicts due to changes to repo README files.	-2
	Request Description is adequate.	-1
TOTAL		-7

Marked as per the due date and changes after which aren't necessarily allowed to contribute to grade for fairness.
Subject to approval from Shakes

[gayanku]

Feedback marks possible +2 if the requested changes are made (see above).

[richardchantra]

Done, thanks team!

[shakes76]

Cant merge because of conflicting changes to main repo files (LICENSE). Please update for merge, doesn't affect grade.