Final submission for OASIS 2D U-Net segmentation project by tmthyngyn · Pull Request #280 · shakes76/PatternAnalysis-2025

Commits on Nov 1, 2025

Init oasis project structure
tmthyngyn
committed

Commits on Nov 2, 2025

Add baseline 2D UNet architecture for OASIS
tmthyngyn
committed
Implement OASIS dataset loader with Colab/Rangpur autodectect
tmthyngyn
committed
Add training driver (CE loss + Dice metric + checkpoint saving)
tmthyngyn
committed
Add prediction script with example visualsation save
tmthyngyn
committed
Add one-hot encoding helper
tmthyngyn
committed
Update OASIS dataset loader to use real Rangpur PNG structure and retain Colab compatibilitiy
tmthyngyn
committed
Support Colab style directory layout (train/image, train/labels)
tmthyngyn
committed
Normalise input image and remap segmentation labels to 0..num_classes-1
tmthyngyn
committed
Update glob pattern to detect .nii.png slices in OASIS dataset
tmthyngyn
committed
Remove debug prints and polish output messages
tmthyngyn
committed
Extend dataset loader with NIfTI option, plotting and class-weight helper
tmthyngyn
committed
Add NIfTI loading utilities, dice helper and visualisation
tmthyngyn
committed
Add CLI training script with validation, class weights, and best-model saving
tmthyngyn
committed
Refactor training script for robustness and compatibility
tmthyngyn
committed
Make predict.py load checkpoints saved by new training script
tmthyngyn
committed
Save prediction visualisation as outputs/prediction_example.png
tmthyngyn
committed

Commits on Nov 3, 2025

Commits on Nov 4, 2025

Commits on Nov 5, 2025

Use correct image
tmthyngyn
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Final submission for OASIS 2D U-Net segmentation project#280

Final submission for OASIS 2D U-Net segmentation project#280
tmthyngyn wants to merge 27 commits intoshakes76:topic-recognitionfrom
tmthyngyn:topic-recognition

Commits on Nov 1, 2025

Init oasis project structure

Commits on Nov 2, 2025

Add baseline 2D UNet architecture for OASIS

Implement OASIS dataset loader with Colab/Rangpur autodectect

Add training driver (CE loss + Dice metric + checkpoint saving)

Add prediction script with example visualsation save

Add one-hot encoding helper

Update OASIS dataset loader to use real Rangpur PNG structure and retain Colab compatibilitiy

Support Colab style directory layout (train/image, train/labels)

Normalise input image and remap segmentation labels to 0..num_classes-1

Update glob pattern to detect .nii.png slices in OASIS dataset

Remove debug prints and polish output messages

Extend dataset loader with NIfTI option, plotting and class-weight helper

Add NIfTI loading utilities, dice helper and visualisation

Add CLI training script with validation, class weights, and best-model saving

Refactor training script for robustness and compatibility

Make predict.py load checkpoints saved by new training script

Save prediction visualisation as outputs/prediction_example.png

Commits on Nov 3, 2025

Enforce canonical data directory structure and unify data loader

Update to fail unless data structure exist

Write inline comments and docstrings

Reading Nifti Files was not needed

Compute Dice score per segmentation

Generate figures for best/worst/median

Commits on Nov 4, 2025

Document final OASIS U-Net results and model performance analysis

Include diagrams and figures in report

Minor changes to fix formatting of report

Commits on Nov 5, 2025

Use correct image

Final submission for OASIS 2D U-Net segmentation project#280tmthyngyn wants to merge 27 commits intoshakes76:topic-recognitionshakes76/PatternAnalysis-2025:topic-recognitionfrom tmthyngyn:topic-recognitiontmthyngyn/PatternAnalysis-2025:topic-recognitionCopy head branch name to clipboard

Commits

Commits on Nov 1, 2025

Commits on Nov 2, 2025

Commits on Nov 3, 2025

Commits on Nov 4, 2025

Commits on Nov 5, 2025

Final submission for OASIS 2D U-Net segmentation project#280
tmthyngyn wants to merge 27 commits intoshakes76:topic-recognitionfrom
tmthyngyn:topic-recognition