SelectionConv: Convolutional Neural Networks for Non-rectilinear Image Data
This work addresses the challenge of processing irregular image domains for computer vision applications, offering a more generalizable approach compared to specialized networks, though it is incremental in building on existing CNN methods.
The paper tackles the problem of applying convolutional neural networks to non-reectilinear image data like spherical images and superpixels by introducing a structured graph convolution operator that transfers weights from pre-trained 2D CNNs, enabling tasks such as segmentation, stylization, and depth prediction without large domain-specific datasets.
Convolutional Neural Networks have revolutionized vision applications. There are image domains and representations, however, that cannot be handled by standard CNNs (e.g., spherical images, superpixels). Such data are usually processed using networks and algorithms specialized for each type. In this work, we show that it may not always be necessary to use specialized neural networks to operate on such spaces. Instead, we introduce a new structured graph convolution operator that can copy 2D convolution weights, transferring the capabilities of already trained traditional CNNs to our new graph network. This network can then operate on any data that can be represented as a positional graph. By converting non-rectilinear data to a graph, we can apply these convolutions on these irregular image domains without requiring training on large domain-specific datasets. Results of transferring pre-trained image networks for segmentation, stylization, and depth prediction are demonstrated for a variety of such data forms.