Published on:
25 April 2024
Primary Category:
Computer Vision and Pattern Recognition
Paper Authors:
Zeynep Özdemir,
Hacer Yalim Keles,
Ömer Özgür Tanrıöver
Examines episodic vs traditional training for rare skin disease classification
Tests models using DenseNet121, MobileNetV2, ImageNet pretraining
Transfer learning represents features well and improves performance
Traditional training surpasses episodic as shots increase
Data augmentation further boosts accuracy
Exploring transfer learning to classify rare skin diseases
This paper examines combining few-shot learning and transfer learning to address challenges in classifying rare skin diseases with limited data. Models using DenseNet121, MobileNetV2, ImageNet pretraining, episodic training, and data augmentation are tested on datasets like SD-198. Key findings show transfer learning helps represent features and boosts performance, traditional training surpasses episodic past few shots, and data augmentation further improves accuracy, achieving state-of-the-art on some datasets.
No comments yet, be the first to start the conversation...
Sign up to comment on this paper