Skin Lesion Classifier

This repository, Skin Lesion Classifier, provides tools for preprocessing, splitting, augmenting, and visualizing image datasets for machine learning tasks, specifically designed for skin lesion classification using the HAM10000 dataset.

Features

• Dataset Splitting: Automatically splits datasets into training, validation, and test sets with proportional representation by class.
• Image Augmentation: Enhances dataset diversity with transformations such as rotation, flipping, cropping, color jitter, and noise addition.
• Visualization: Generates bar charts, heatmaps, and sample image displays for dataset analysis.

File Overview

• augmentation_to_image.py: Script for applying image augmentation to the training set and saving the results.
• dataset_visualization.ipynb: Notebook for dataset preprocessing, splitting, visualization, and sample image display.
• .gitignore: Excludes unnecessary files and directories from version control.

How to Use

0. Clone the Repository and Download the Dataset

• Clone the repository:

  git clone https://github.com/zxia545/SkinLesionClassifier.git  

• Download the HAM10000 dataset from Kaggle.

1. Dataset Splitting

• Organize your dataset into dataset/ with subfolders for each class.
• Run dataset_visualization.ipynb to split the dataset into training, validation, and test sets.

2. Image Augmentation

• Augment the training data by running:

  python augmentation_to_image.py  

• Augmented images will be saved in split_data/augmented_train/.

3. Visualization

• Use dataset_visualization.ipynb to:
  • Generate bar charts and heatmaps for class distribution.
  • Display random sample images from training, validation, and test sets.

4. Model Training

• Use main.py (if present) to train models. Adjust dataset paths in the script to match your directory structure.

5. Evaluation

• Evaluate models using load_and_test.py, ensuring it aligns with your trained model and test dataset structure.

Dependencies

Install required Python libraries using:

pip install -r requirements.txt  

Directory Structure

SkinLesionClassifier/
├── augmentation_to_image.py
├── dataset_visualization.ipynb
├── .gitignore
├── requirements.txt
├── split_data/
│   ├── train/
│   ├── validation/
│   ├── test/
│   ├── augmented_train/
└── dataset/
    └── Labeled Data/

Notes

• Modify paths in scripts as necessary for your environment.
• Ensure datasets are organized with subfolders for each class.
• For vision-language model experiments (CLIP, BLIP), include appropriate pre-trained weights and scripts.

License

This project is licensed under the MIT License.

Acknowledgments

This repository supports the research in "Exploring Vision-Language and Deep Learning Models for Skin Cancer Classification" by Zheyuan Xiao, utilizing HAM10000 for experimentation with methods like Random Forest, ResNet, Custom CNNs, and vision-language models (CLIP, BLIP).