DerMAE: Improving skin lesion classification through conditioned latent diffusion and MAE distillation

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Filho, Francisco, Cunha, Kelvin, Papais, Fábio, Santos, Emanoel dos, Mota, Rodrigo, Bezerra, Thales, Medeiros, Erico, Borba, Paulo, Ren, Tsang Ing
Formato: Preprint
Publicado: 2026
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866911740674965504
author Filho, Francisco
Cunha, Kelvin
Papais, Fábio
Santos, Emanoel dos
Mota, Rodrigo
Bezerra, Thales
Medeiros, Erico
Borba, Paulo
Ren, Tsang Ing
author_facet Filho, Francisco
Cunha, Kelvin
Papais, Fábio
Santos, Emanoel dos
Mota, Rodrigo
Bezerra, Thales
Medeiros, Erico
Borba, Paulo
Ren, Tsang Ing
contents Skin lesion classification datasets often suffer from severe class imbalance, with malignant cases significantly underrepresented, leading to biased decision boundaries during deep learning training. We address this challenge using class-conditioned diffusion models to generate synthetic dermatological images, followed by self-supervised MAE pretraining to enable huge ViT models to learn robust, domain-relevant features. To support deployment in practical clinical settings, where lightweight models are required, we apply knowledge distillation to transfer these representations to a smaller ViT student suitable for mobile devices. Our results show that MAE pretraining on synthetic data, combined with distillation, improves classification performance while enabling efficient on-device inference for practical clinical use.
format Preprint
id arxiv_https___arxiv_org_abs_2602_19848
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle DerMAE: Improving skin lesion classification through conditioned latent diffusion and MAE distillation
Filho, Francisco
Cunha, Kelvin
Papais, Fábio
Santos, Emanoel dos
Mota, Rodrigo
Bezerra, Thales
Medeiros, Erico
Borba, Paulo
Ren, Tsang Ing
Computer Vision and Pattern Recognition
Skin lesion classification datasets often suffer from severe class imbalance, with malignant cases significantly underrepresented, leading to biased decision boundaries during deep learning training. We address this challenge using class-conditioned diffusion models to generate synthetic dermatological images, followed by self-supervised MAE pretraining to enable huge ViT models to learn robust, domain-relevant features. To support deployment in practical clinical settings, where lightweight models are required, we apply knowledge distillation to transfer these representations to a smaller ViT student suitable for mobile devices. Our results show that MAE pretraining on synthetic data, combined with distillation, improves classification performance while enabling efficient on-device inference for practical clinical use.
title DerMAE: Improving skin lesion classification through conditioned latent diffusion and MAE distillation
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2602.19848