Do not smile: Selfie Skin Profiling

This repository contains the baseline code, documentation, and evaluation scripts for the paper: "Do not smile: Selfie Skin Profiling using DeepLearning", published in Scientific Reports.

1. Overview

This project presents a baseline pipeline for automated skin feature profiling based on unconstrained consumer imagery. It serves to provide transparency into our methodology and the evaluation metrics used for the research.

2. Data Availability

The raw dataset is not available for distribution. Due to strict General Data Protection Regulation (GDPR) compliance and our commitment to guaranteeing complete consumer anonymity, the underlying image dataset cannot be shared or hosted publicly. Our methodology relied on decentralized acquisition without the collection of identifiable personal metadata, and we maintain these strict privacy constraints to protect the participants.

3. Methodology & Baseline

The repository includes the baseline pipeline used to generate the results presented in our study.

Prerequisites

• Python 3.10
• pip install -r requirements.txt
• curl -L -o models/face_detection_yunet_2023mar.onnx " "https://github.com/opencv/opencv_zoo/raw/main/models/face_detection_yunet/" "face_detection_yunet_2023mar.onnx

Pipeline

The pipeline consists of:

1. Preprocessing: Facial region extraction using standard open-source face analysis libraries (target resolution: 224x224).
2. Model Architecture: [Insert your model architecture, e.g., ResNet/EfficientNet/Custom].
3. Evaluation: Metrics for ordinally scaled attributes using Spearman’s rank correlation and Weighted Kappa to account for distributional skew.

4. Ethical Considerations & Limitations

• Demographic Composition: The dataset is predominantly composed of Caucasian skin profiles.
• Privacy-by-Design: Self-reported demographic metadata (sex, ethnicity) was not collected to guarantee consumer anonymity. Post-hoc demographic estimations provided in the paper are subjective and should be used with caution regarding generalization.
• Disclaimer: This system is strictly designed as an appearance-based profiling tool for skincare guidance and is not a medical diagnostic tool.

5. Reproducibility

The code provided allows for the reproduction of our baseline performance metrics. Change preprocessing.py, run.py and evaluation.py according to your file structure.

python starter.py

6. Citation

If you utilize this code or the methodology described in our work, please cite our paper: Code-Snippet

@article{Hartmann2026,
  title={Do not smile: Selfie Skin Profiling using DeepLearning},
  author={Hartmann, Dennis et al.},
  journal={Scientific Reports},
  year={2026},
  publisher={Nature Portfolio}
}

7. License

Code: This repository's source code is released under the Apache License.

Data: The dataset is not included in this repository and is strictly unavailable for public release or distribution due to privacy and GDPR constraints.