YOLO Detection of Lesions in ISIC Dataset

[mermalade0325]

#This pull request aims to address the problem of detecting skin lesions within the ISIC 2017/2018 dataset. The primary goal is to develop a robust detection and classification network that can accurately identify lesions in medical images, ensuring each detection has a minimum Intersection Over Union (IoU) of 0.8 on the test set, thus achieving a suitable level of accuracy for classification. The project utilizes a YOLO-based detection network (YOLOv7).

Problem Statement
Skin cancer detection relies on identifying lesions accurately in medical images. This project contributes by building a deep learning model focused on detecting lesion locations without classifying lesion types. The main goals are:

• Localization: Detect lesion locations with high accuracy.
• Performance Metrics: Achieve a minimum IoU of 0.8 for lesion detection on the test set.

Files and Structure

The project is organized into the following files:

• modules.py: Contains core components of the model, structured as classes or functions for easy reuse and extension.
• dataset.py: Implements the data loader for the ISIC dataset, with preprocessing functions.
• train.py: Contains training, validation, and testing routines. It imports the model from modules.py and data loader from dataset.py. Training metrics, such as loss, are plotted and saved to help track model performance.
• predict.py: Shows how to load the trained model for predictions. Outputs include visualizations with bounding boxes around detected lesions.
• README.md: Documents the project setup, usage, and module descriptions.

Dataset

Regarding the data downloaded from ISIC, it should be organised as follows, with labels generated from dataset.py:

An appendix in the README file links to the ISIC dataset for reference.

Correctly merged pull request: #193 (comment)

[shaivikaaaa]

@gayanku

[shaivikaaaa]

This is an initial inspection

Difficulty : Normal
Task : 3

1. Recognition Problem:

• file structure is ok
• Include train, test , validation % in PR and README
• no plots/images on classification results shown
• minimal comments
• no proof of training and testing
• model trained for 10 epochs

2. Commit Log:

• Commit message not good, or progressive
• Commit messages are only related to README file

3. Documentation:
   Tittle: Done
   Description of the algorithm: Done
   Problem that it solves: Done
   How it works in a paragraph (example any pre-processing): Done
   Figure/visualisation (any input, output & results): Not done
   Dependencies: done

4. Pull Request
   Elaborate about problem you are solving, and what you have done - a short description

Feedback:

• Update PR a bit
• If you have trained and tested images - upload those

[aniketgupta17]

Observational Feedback

Pull Request:
Incorrectly created the Pull request from main Branch .
The pull request should include a clear description about the file structure .
No previous feedback incorporated .

File Organizing: Incorrect way , See the proper organisation files in the task sheet .

Commit Log:
Commit messages are not progressive for the Recognition Problem solved using 4 files .
Commits are not regularly made showing no logical development.

Documentation:
Code comments and docstrings are included.
Proper GitHub markdown formatting is used, with organized headings, lists, and code blocks.
Should also add discussion about the testing of the model .
Should add about future improvements in the ReadMe.

[mermalade0325]

Hi all,I received the first feedback on Saturday evening and I have been in the process of implementing it all day yesterday and today. Please let me know when the final decision for marking will be made, as a comment on Ed by tutors mentioned it would be after a week since the first feedback.Thank you so much for the advice and support. I have had a number of challenges this year in terms of health reasons and I appreciate the opportunity to implement the feedback. Kind regards, Mariam

[mermalade0325]

#193 (comment)

Hi team,
As requested by tutors I have made updates with their feedback - reorganising the files correctly and submitting a new, correct, pull request from the topic-recognition branch of my repository.

I have attached the link above. Please consider this pull request for my project. I will continue to work on it to implement your feedback, as per Ed's mention of the final date being the end of exam block. If requested I can remove the old files which were incorrectly included in the main branch but for now I will keep them there in case they are required as proof of timely submission dates.

Thank you for the opportunity to implement your feedback.

Kind regards,
Mariam

[shaivikaaaa]

Hi @mermalade0325 that sounds fine
If you have incorporated the required feedback then that is what we need

[gayanku]

Marking

Good/OK/Fair Practice (Design/Commenting, TF/Torch Usage)
	Good design and implementation.
	Spacing and comments.
	No Header blocks.	-1
Recognition Problem
	Good solution to problem.
	Driver Script present.
	File structure NOT present.	-1
	Good Usage & Demo & Visualisation & Data usage.
	Module present.
	Commenting present.
	No Data leakage found. Has Data splits.
	Difficulty : Normal. YOLO (Normal Difficulty)	-5
Commit Log
	Some/Adequate Meaningful commit messages.	-1
	Good Progressive commits.	-2
Documentation
	Readme :Acceptable. Incomplete. Need train curves etc.	-2
	Model/technical explanation :Good.
	Description and Comments :Good.
	Markdown used and PDF submitted. PDF Checked.
Pull Request
	Pull Request has problems: Late submission. Incorrect PR from main.	-2
	Feedback action required: Feedback marks possible +2 if the requested changes are made. Correct the branch. Do NOT create a new PR, instead correct this PR for merge.	-2
	Request Description is adequate.	-1
TOTAL		-17

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).

[shakes76]

Approved extension and fixed PR in #193 +2