Melanoma dataset

This repository contains a curated, multi-source dataset specifically structured to benchmark machine learning models on automated skin lesion classification. The data aggregates several major open-access medical imaging datasets, offering a robust blend of high-magnification dermoscopic images (ideal clinical baseline) and clinical smartphone macro photography (real-world, noisy consumer testing). By combining these files under a single pipeline framework, this dataset provides a comprehensive platform for testing model generalization, dataset harmonization, and evaluating performance drops caused by real-world domain shifts. Data Provenance & Sources The repository unifies data from the following authoritative medical datasets: ISIC 2016 Archive: Standard dermoscopic images focused on binary classification benchmarks. ISIC 2020 Archive: Large-scale, high-resolution dermoscopic imagery representing highly diverse lesion morphologies. PAD-UFES-20: A specialized clinical dataset collected via smartphone cameras (primarily Samsung Galaxy hardware) in university dermatological assistance programs, subject to variable domestic lighting conditions, framing distances, and skin background noise. Kaggle Components: Supplemental benchmark tracking distributions.