We investigate the problem of training neural networks from incomplete images without replacing missing values. For this purpose, we first represent an image as a graph, in which missing pixels are entirely ignored. The graph image representation is processed using a spatial graph convolutional network (SGCN) -- a type of graph convolutional networks, which is a proper generalization of classical CNNs operating on images. On one hand, our approach avoids the problem of missing data imputation while, on the other hand, there is a natural correspondence between CNNs and SGCN. Experiments confirm that our approach performs better than analogical CNNs with the imputation of missing values on typical classification and reconstruction tasks.
@article{arxiv.2010.13914,
title = {Processing of incomplete images by (graph) convolutional neural networks},
author = {Tomasz Danel and Marek Śmieja and Łukasz Struski and Przemysław Spurek and Łukasz Maziarka},
journal= {arXiv preprint arXiv:2010.13914},
year = {2020}
}