RuiqiYang77/skin-svm

Skin Lesion SVM Dataset

This dataset contains RGB skin lesion images, binary lesion masks, and class labels for a three-class biomedical image classification task. It is designed for evaluating classical image-processing features and machine-learning models, especially the SVM baseline implemented in the companion code repository.

Code repository: https://github.com/RuiqiYang77/skin-svm

Dataset Structure

The dataset contains only two splits: train and test.

data/
|-- train/
|   |-- image/
|   |-- mask/
|   `-- label.csv
`-- test/
    |-- image/
    |-- mask/
    `-- label.csv

Each split follows the same format:

• image/: RGB lesion images in .jpg format.
• mask/: binary lesion masks in .jpg format.
• label.csv: image-level class labels.

Masks are named with the mask_ prefix. For example:

image/5696.jpg
mask/mask_5696.jpg

Labels

The task contains three diagnostic categories:

• nv: melanocytic nevus
• mel: melanoma
• vasc: vascular lesion

Each label.csv file contains:

image_id,dx

where:

• image_id is the image filename without the .jpg suffix.
• dx is the class label.

Split Statistics

Split	Total	nv	mel	vasc
train	584	200	200	184
test	150	50	50	50

Intended Use

This dataset is intended for research and coursework on biomedical image processing, handcrafted feature extraction, and robust evaluation of classical machine-learning models. The companion repository provides scripts for:

• feature extraction from images and masks
• SVM training
• external test evaluation
• single-image prediction
• robustness testing under light image noise

Loading Example

import pandas as pd
from pathlib import Path

root = Path("data")
train_labels = pd.read_csv(root / "train" / "label.csv")

row = train_labels.iloc[0]
image_path = root / "train" / "image" / f"{row.image_id}.jpg"
mask_path = root / "train" / "mask" / f"mask_{row.image_id}.jpg"

print(row.dx, image_path, mask_path)