s43971125 #291

Alex605268 · 2025-11-07T05:54:48Z

Alexander Renkin-Perry
s43971125
Implementation of Task 1: Segmentation of 2D_OASIS dataset using improved UNet.
Adds 4 required python files, README and a documentation folder that contains 5 images used in the README

Train.py by default creates a model titled "improved_best.pth". predict.py expects that file to be in the same directory when it runs.
Train.py contains print statements to check the models process, they have been commented out. Predict.py saves all the plots
it generates to pngs in the same folder.

…le.py. Implemented very basic UNet model in module.py, including a test script for a forward pass without any data to ensure that the input will pass through all layers without error Updated README.md to list all the current files used in the project

Added functions for loading 2D data provided by assignment - Realised after that they aren't necessary for the OASIS dataset, but as I hope to tackle the HipMRI dataset if I can complete this task, I have left them in for now. Will remove before submission if they are still unused.

…comments - more needed. Added get_dataloaders function to dataset.py. Added DiceLoss coefficient calculation functionality to train.py - currently only works for binary segmentation - may need to be changed.

…rected functions, missing variable initialisations). Tested to find that 2D_OASIS data has 4 segmentation classes, so updated DiceLoss functionality accordingly Ran a quick test with one epoch, and confirmed that it created a "best.pth" file with no errors - now ready to test with multiple epochs, adjusted parameters, etc

…op works Added functionality to print out results of each epoch to log file. Fixed a few minor errors. Making this commit before running a longer test to see if results improve between training loops.

Previously it was calibrated for Binary segmentation, so it gave a totally incorrect answer. Should be working correctly now.

… run train.py Existed beforehand, forgot to commit it earlier. Not many changes here - attempted to get HIPMri working but was having issues loading the data, eventually just reverted to the 2D_OASIS, so eventually just reverted back to what I had before, with a couple of improvements made to check progress during training

Currently loads model and one image, does a prediction for that image, and visualises results May update again batch over all test images. Only changes to train.py are in the parameters (batch size and epoch number)

Adds some comments to modules.py explaining improved UNet implementation - more to come Adjusted predict.py functionality to export visualisation to png Adjusted parameters in train.py

Adds skeleton to README.md that includes all potential Headers and hyperlinks

…will be linked to in the README file. Adds information about the problem and model to the READme file. Adjusts the parameters of train.py.

Updates README with information about training the model

…he model and the fuction that provides visualisation. Adds a second visualisation function that shows the per-class visualisation

…scores

…ection Final adjustment to train.py so default parameters match those in README Fixed the function in predict.py to properly graph the dice scores Adds the image of the dice scores to the documentation

wangzhaomxy · 2025-11-21T19:48:26Z

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

File Organizing: Well-organized files.

Problem Solving:

The algorithm solves the problem appropriately.
Accuracy in testing dataset (Dice): min 0.95. However, the evidence figure failed to display.

Model and functions:

It correctly uses PyTorch to construct the improved UNet 2D models and functions.
NO data augmentation.
Properly use the train/validation/test datasets.

Code design: Good.

Code comment and docstring:

Good code comments
Good function docstrings
Good header block

Difficulty: Easy.

Additional Comments:

Minimal commits, consisting mostly of generic messages such as ‘update’ or ‘add function.’ More meaningful and descriptive commit messages are expected.
The README design can be more structured and provide more comprehensive content. Since this is a report, a proper Discussion and Conclusion section is also expected to clearly summarize your project.
A training/validation curve is expected to illustrate and monitor the training process.

gayanku · 2025-11-24T05:55:47Z

Marking

Good/OK/Fair Practice (Design/Commenting, TF/Torch Usage)
	Good design and implementation.
	Spacing and comments.
	Header blocks.
Recognition Problem
	Good solution to problem.
	Driver Script present.
	File structure present.
	Good Usage & Demo & Visualisation & Data usage.
	Module present.
	Commenting present.
	No Data leakage found.
	Difficulty : Easy. Easy. ImprovedUnet2D	-10
Commit Log
	Good Meaningful commit messages.
	Good Progressive commits.
Documentation
	Readme :Acceptable.	-2
	Model/technical explanation :Good.
	Description and Comments :Good.
	Markdown used and PDF submitted.
Pull Request
	Successful Pull Request (Working Algorithm Delivered on Time in Correct Branch).
	No Feedback required.
	Request Description is good.
TOTAL		-12

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

s4397112 added 16 commits October 20, 2025 18:32

Filled out train.py with the basic training functionality. Adds some …

010fc12

…comments - more needed. Added get_dataloaders function to dataset.py. Added DiceLoss coefficient calculation functionality to train.py - currently only works for binary segmentation - may need to be changed.

A few more changes to dataset and train. Established that training lo…

a5f22f7

…op works Added functionality to print out results of each epoch to log file. Fixed a few minor errors. Making this commit before running a longer test to see if results improve between training loops.

Updates the dice_coefficient function to match the DiceLoss function

a638d83

Previously it was calibrated for Binary segmentation, so it gave a totally incorrect answer. Should be working correctly now.

Filled out predict.py to do a very simple prediction

9d4577d

Currently loads model and one image, does a prediction for that image, and visualises results May update again batch over all test images. Only changes to train.py are in the parameters (batch size and epoch number)

Adds Improved UNet model to modules.py

b886db6

Adds some comments to modules.py explaining improved UNet implementation - more to come Adjusted predict.py functionality to export visualisation to png Adjusted parameters in train.py

Adds comments to modules, train, dataset.

2a46226

Adds skeleton to README.md that includes all potential Headers and hyperlinks

Add documentation folder. This contains all the relevant images that …

930c9d6

…will be linked to in the README file. Adds information about the problem and model to the READme file. Adjusts the parameters of train.py.

Adds some comments to modules.py

2e55dbc

Updates README with information about training the model

Updates predict.py to be modular (separates the function that loads t…

faadaab

…he model and the fuction that provides visualisation. Adds a second visualisation function that shows the per-class visualisation

Add a third visualisation function to predict.py that plots the dice …

0c7e3d0

…scores

Completion of README file, adds testing, results, and documentation s…

0c5b5ec

…ection Final adjustment to train.py so default parameters match those in README Fixed the function in predict.py to properly graph the dice scores Adds the image of the dice scores to the documentation

Removed slurm script from Repo

3b5998e

hanemma7moud added _Easy _ImprovedUnet2D _Normal and removed _Easy _Normal labels Nov 10, 2025

wangzhaomxy added checked-1st Uploaded_PDF labels Nov 21, 2025

gayanku added the SECOND_MARK label Nov 24, 2025

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

s43971125 #291

s43971125 #291

Uh oh!

Alex605268 commented Nov 7, 2025

Uh oh!

wangzhaomxy commented Nov 21, 2025 •

edited

Loading

Uh oh!

gayanku commented Nov 24, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

s43971125 #291

Are you sure you want to change the base?

s43971125 #291

Uh oh!

Conversation

Alex605268 commented Nov 7, 2025

Uh oh!

wangzhaomxy commented Nov 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

gayanku commented Nov 24, 2025

Marking

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

wangzhaomxy commented Nov 21, 2025 •

edited

Loading