Beyond Blur: A Semantic Tri-view Pipeline for Teledermatology Gradability via Skin Micro-relief

The paper introduces the Semantic Tri-view Pipeline, an interpretable system that automatically screens teledermatology photographs for gradability by analyzing epidermal micro-relief across up to three smartphone views. It uses a lightweight DeepLabV3+ model to segment micro-relief fidelity and aggregates the resulting spatial masks with logistic regression, leveraging viewpoint redundancy to improve robustness. Evaluated on the SCIN dataset, the approach raises the AUC from 0.81 to 0.96 on optically clear cases, offering real‑time, privacy‑by‑design feedback to filter ungradable photo sets before clinician review.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Machine Learning.

• safety