CORTEXA
← Browse
openalexSignal Image and Video Processing2026-07-25Cited by 0

Beyond nearest neighbor interpolation in data augmentation

Olivier Rukundo

Abstract Avoiding the risk of undefined categorical labels using nearest-neighbor interpolation overlooks the risk of exacerbating pixel-level annotation errors in augmented training data. Additionally, the inherent low-pass filtering effects of interpolation algorithms exacerbate the risk of degrading high-frequency structural details within annotated regions of interest. To avoid these risks, the author modified convolutional neural networks’ data transformation functions by incorporating a modified geometric transformation function, removing reliance on nearest-neighbor interpolation, and integrating a mean-based class filtering mechanism to handle undefined categorical labels with alternative interpolation algorithms. The author also implemented an offline data augmentation pipeline to generate interpolation-specific augmented training data, enabling quantitative assessment of interpolation-specific low-pass filtering effects on augmented training data. Experimental evaluation on three medical image segmentation datasets and the XBAT + datasets demonstrated performance gains across multiple quantitative metrics.

View free PDFSource page

Related papers

openalexSignal Image and Video Processing2026-07-24

Automated fault detection in solar panels using customized EfficientNetB0 with explainable AI solution for real-time monitoring

S Sneha, Anik Sen, Sumaiya Malik, M Kamal Hossen

Solar energy offers a sustainable solution for power generation, reducing dependence on fossil fuels. However, surface-level anomalies such as dust, snow, bird droppings, and structural damage significantly impact the operational efficiency of solar panels. This study presents a…

View free PDFSource page