Acne Detection & Severity Grading with YOLOv8 (ICIIT 2024)

This repository provides a reproducible pipeline for acne lesion detection and skin damage severity grading (IGA scale 0–4) using YOLOv8.

Paper: Initial Approach to the Identification Degree of Skin Damage and Classification of Acne by YOLOv8 (ICIIT 2024).
DOI: 10.1145/3654522.3654526

---

1. Overview

Acne is a common dermatological condition and manual assessment can be time-consuming and subjective. This project uses YOLOv8 for object detection of acne lesion types and maps the detected outcomes to acne severity grades according to the IGA scale (0–4).

Target lesion classes (5):

• acne scars
• blackheads
• papular
• purulent (pustules)
• sebo-crystan-conglo

---

2. Dataset

• Source: DermNet-NZ (annotated and prepared via Roboflow in our experiments)
• Total images: 566
• Image size used for training: 640×640

Class distribution (images / objects depend on annotation setup):

• Acne scars: 119
• Blackheads: 97
• Papular: 121
• Purulent: 114
• Sebo-crystan-conglo: 115
• Total: 566

⚠️ Dataset files are not included in this repository.
Please follow data/README_DATA.md to obtain and prepare data, and respect the original dataset license/terms.

---

3. Environment

Our reference training environment:

• OS: Ubuntu 20.04.6 LTS
• GPU: Tesla T4 (16 GB)
• CUDA: 12.0
• Framework: PyTorch
• Python: 3.10
• Training workflow: Google Colab + Anaconda

---

4. Installation (YOLOv8 / Ultralytics)

# (Recommended) Create a virtual environment
python -m venv .venv
# Windows: .venv\Scripts\activate
source .venv/bin/activate

pip install --upgrade pip
pip install ultralytics