Neural Multigrid

We propose a multigrid extension of convolutional neural networks (CNNs). Rather than manipulating representations living on a single spatial grid, our network layers operate across scale space, on a pyramid of tensors. They consume multigrid inputs and produce multigrid outputs; convolutional filters themselves have both within-scale and cross-scale extent. This aspect is distinct from simple multiscale designs, which only process the input at different scales. Viewed in terms of information flow, a multigrid network passes messages across a spatial pyramid. As a consequence, receptive field size grows exponentially with depth, facilitating rapid integration of context. Most critically, multigrid structure enables networks to learn internal attention and dynamic routing mechanisms, and use them to accomplish tasks on which modern CNNs fail. Experiments demonstrate wide-ranging performance advantages of multigrid. On CIFAR image classification, flipping from single to multigrid within standard CNN architectures improves accuracy at modest compute and parameter increase. Multigrid is independent of other architectural choices; we show synergistic results in combination with residual connections. On tasks demanding per-pixel output, gains can be substantial. We show dramatic improvement on a synthetic semantic segmentation dataset. Strikingly, we show that relatively shallow multigrid networks can learn to directly perform spatial transformation tasks, where, in contrast, current CNNs fail. Together, our results suggest that continuous evolution of features on a multigrid pyramid could replace virtually all existing CNN designs.