makhresearch/skin-lesion-segmentation-classification

🔗 Usage with 🤗 Datasets Library

# ==============================================================================
# Final and reliable method — clean dataset structure, no .cast() required
# ==============================================================================

# Step 1: Install the Hugging Face datasets library
!pip install datasets -q

# Step 2: Download and unzip the dataset (recommended method)
import requests
from zipfile import ZipFile
from io import BytesIO
from pathlib import Path

url = "https://huggingface.co/datasets/makhresearch/skin-lesion-segmentation-classification/resolve/main/skin-lesion-segmentation-classification.zip"
print("Downloading the dataset ZIP file...")
response = requests.get(url)
response.raise_for_status()
print("✅ Download complete.")

print("\nExtracting files...")
with ZipFile(BytesIO(response.content)) as zf:
    zf.extractall(".")
print("✅✅✅ Extraction complete.")

# ==============================================================================
# Done
# ==============================================================================

---

🧠✨ Skin Lesion Segmentation & Classification Dataset

Welcome to Skin Lesion Segmentation & Classification — a high-quality medical dataset 🧬 focused on dermatological image analysis using bounding boxes and segmentation annotations. Whether you're training a model for melanoma detection or experimenting with computer vision in healthcare, this dataset is built for you! 🚀🩺

---

📦 Dataset Overview

🔍 This dataset follows the YOLO-style folder structure, ready for training in detection or segmentation tasks:

skin-lesion-segmentation-classification/
├── train/
│   ├── images/     # 6,675 training images
│   └── labels/     # YOLO format labels
├── valid/
│   ├── images/     # 1,911 validation images
│   └── labels/
├── test/
│   ├── images/     # 961 test images
│   └── labels/

📄 Each .txt label contains YOLO format:

<class_id> x1 y1 x2 y2 x3 y3 ... xn yn

✅ All coordinates are normalized between 0 and 1.

---

🧠 Classes

The dataset supports 7 lesion categories, labeled from ID 0 to 6. Below is a full list of class codes, descriptions, and emojis for quick visual reference:

🦠 Skin Lesion Types & Description

ID	Emoji	Code	Full Name	Description
0	🟤	BKL	Benign Keratosis	Non-cancerous, often scaly skin lesions. Common and usually harmless.
1	⚫	NV	Melanocytic Nevi	Regular moles formed by pigment-producing cells. Typically benign.
2	🟠	DF	Dermatofibroma	Firm, small nodules under the skin caused by minor trauma. Non-cancerous.
3	🔴	MEL	Melanoma	A serious and potentially life-threatening skin cancer. Early detection is critical.
4	🔵	VASC	Vascular Lesion	Blood vessel-related marks like angiomas. Usually red or purple.
5	🟣	BCC	Basal Cell Carcinoma	The most common skin cancer. Slow-growing and rarely metastasizes.
6	⚠️	AKIEC	Actinic Keratoses / Intraepithelial Carcinoma	Pre-cancerous lesions that may evolve into squamous cell carcinoma.

---

🧪 Tasks

• 🔳 Object Detection (YOLO format with normalized bounding boxes)
• 🎯 Segmentation Approximation (using bounding box areas)
• 📌 Potential for classification/fine-tuning tasks

---

📦 Dataset Size

Split	Images	Labels
Train	6,675	6,675
Valid	1,911	1,911
Test	961	961
Total	9,547	9,547

---

📝 Notes

• All lesion images are in .jpg format.
• Labels are compatible with YOLOv8 training pipelines.
• The dataset is suitable for medical AI, research, and deep learning education.

---

🧠 Use Cases

This dataset can power a wide range of AI tasks:

• 🧪 Skin lesion detection
• 🩻 Medical image segmentation
• 🤖 Disease classification (e.g., melanoma vs benign)
• 📊 Benchmarking object detection models in the medical domain

---

🏷️ Classes

🔢 Supports one or more lesion classes (e.g., lesion).
You can edit the class list based on your labeling config.

---

📜 License

🆓 Open for academic and commercial use under the MIT License.

---

🧾 Citation

If you use this dataset in your research or project, please cite:

@dataset{makhresearch_skin_lesion_2025,
  title     = {Skin Lesion Segmentation and Classification Dataset},
  author    = {MakhResearch},
  year      = {2025},
  url       = {https://huggingface.co/datasets/makhresearch/skin-lesion-segmentation-classification}
}

---

🤝 Contributing

🙌 Contributions, feedback, and ideas are always welcome!

Let’s build better medical AI — together. 💡🩺

---

📬 For questions, citations, or collaboration requests, visit the Hugging Face Profile.