Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Topic recognition pull request - COMP3710 Open Source Project #447

Open
wants to merge 16 commits into
base: topic-recognition
Choose a base branch
from
Open

Topic recognition pull request - COMP3710 Open Source Project #447

wants to merge 16 commits into from

Conversation

bushlemon
Copy link

Implementation of Siamese Network to classify Alzheimer's disease in brain MRIs

@bushlemon
Created boilerplate files and folders
f51ff28 
Created file structure. Files currently empty
@bushlemon
Created image pre-processing
abfbd5d 
Created a function to load the images from a directory and convert to a numpy array
@bushlemon
Defined model generator functions
e0769e3 
- Created the functions necessary for building the NNs that will be used for training and classification.
- Siamese model needs to be tweaked for performance and output size
- Binary classifier needs to be built as it only has the barebones
@bushlemon
Added docstrings to load_data function
31f0215 
Added documentation to load_data function in dataset.py
@bushlemon
Began implementing training code
1767018 
Created train_step function
Various other documentation updates
@bushlemon
Created main method in train.py
c422b40 
Further built out functionality of train.py
Laid groundwork for README
@bushlemon
Further implementation in training pipeline
0a0fa37 
- Moved loss function from modules to train
- Defined training function for siamese
- Fleshed out main function
@bushlemon
Implemented Binary Classifier
cc3adf7 
- Created binary classifier function in modules.py
- Created train_binary_classifier function in train.py
@bushlemon
Defined siamese loss function
b650559 
- Defined siamese loss function
- Other minor changes
@bushlemon
Finished first pass of training
c9530af 
- Got training working for both models
- Model tuning still required
- Starting to fill out predict.py
@bushlemon
Edited Binary Classifier
c043eb9 
- changed structure of binary NN
- changed binary classifier training function to feed in entire dataset rather than pre-batched
@bushlemon
Minor tweaks and documentation additions
f770e8b
@bushlemon
Various readme and documentation updates
fe0d94e 
- Updated documentation
- Added PCA plot
@bushlemon
More additions to documentation
ab60786
@bushlemon
Added principal component analysis to train.py
3aa5429 
- Also further documentation adjustments
@bushlemon
Finished Readme
1bc7659
@nathasha-naranpanawa
Copy link
Collaborator

nathasha-naranpanawa commented Nov 14, 2022 •
edited
Loading

  • Commit contains files from other students. Clean up might be necessary
  • Commit mesages are OK, could be more descriptive
  • Readme is OK, could have been better if architecture diagrams were provided, no training/testing loss or accuracy plots shown, no mention of how the data were preprocessed
  • Performance is low (51% classification accuracy). Siamese doesn't seem to be learning anything useful as shown by the 2D embedding plot.
  • Have tried several workarounds to try and improve performance, which hasn't really worked
  • Looking at the code, seem to be concatenating positive sample embeddings and negative sample embeddings one after another but not shuffle them before feeding them into the binary classifier. This could be one reason the classifier is acting poorly.
  • Code is OK, but classification logic is incorrect - You need to produce embeddings with trained Siamese for individual images, not pairs of images. Then, the labels for binary classification need to be 'have Alzheimer's/not', not whether they are similar or not to each other within a pair.
  • No prediction results shown with example images

@shakes76
Copy link
Owner

Good Practice (Design/Commenting, TF/Torch Usage)

Adequate use and implementation (logic errors in code, but works for the most part) -2
Good spacing and comments
Header blocks missing -1

Recognition Problem

Solves problem (poor performance) -2
Driver Script present
File structure present
Shows Usage & Demo & Visualisation & Data usage (no prediction outputs or plots) -2
Module present
Commenting
No Data leakage
Difficulty: Hard

Commit Log

Meaningful commit messages
Progressive commits used

Documentation

ReadMe OK, no preprocessing info, diagrams etc. -2
Good Description and Comments
Markdown used PDF submitted

Pull Request

Successful Pull Request (Working Algorithm Delivered on Time in Correct Branch)
Feedback required, revert changes to other student files, fix conflict -2
Request Description minimal, needs more info -1

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
Feedback_Needed Siamese Network
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
5 participants
@bushlemon @nathasha-naranpanawa @shakes76 @SiyuLiu0329 @LinfengLiu98