SuperpixelGridMasks Data Augmentation: Application to Precision Health and Other Real-world Data.

Area of Science:

• Computer Vision
• Machine Learning
• Data Augmentation

Background:

• Machine learning models require extensive datasets for optimal performance.
• Existing data augmentation techniques may not fully leverage image data characteristics.

Purpose of the Study:

• To introduce SuperpixelGridMasks, a novel data augmentation approach for enhancing machine learning model performance.
• To present three variants: SuperpixelGridCut, SuperpixelGridMean, and SuperpixelGridMix.

Main Methods:

• Irregular superpixel decomposition for data augmentation.
• Grid-based image transformations involving information dropping and fusing.
• Implementation of SuperpixelGridCut, SuperpixelGridMean, and SuperpixelGridMix variants.

Main Results:

• SuperpixelGridMasks significantly outperformed baseline performances across various image classification models.
• The proposed methods surpassed the performance of other existing data augmentation techniques.
• Demonstrated effectiveness on precision health and real-world datasets.

Conclusions:

• SuperpixelGridMasks offers a powerful new approach to data augmentation in machine learning.
• The method's variants provide versatile tools for improving model accuracy and robustness.
• Publicly available code facilitates adoption and further research.