Blurring Structure and Learning to Optimize and Adapt Receptive Fields

Evan Shelhamer, Dequan Wang, Trevor Darrell

Keywords: scale, deep learning, dynamic inference, fully convolutional

TL;DR: Composing structured Gaussian and free-form filters makes receptive field size and shape differentiable for end-to-end optimization and dynamic adaptation.

Abstract: The visual world is vast and varied, but its variations divide into structured and unstructured factors. We compose free-form filters and structured Gaussian filters, optimized end-to-end, to factorize deep representations and learn both local features and their degree of locality. In effect this optimizes over receptive field size and shape, tuning locality to the data and task. Our semi-structured composition is strictly more expressive than free-form filtering, and changes in its structured parameters would require changes in architecture for standard networks. Dynamic inference, in which the Gaussian structure varies with the input, adapts receptive field size to compensate for local scale variation. Optimizing receptive field size improves semantic segmentation accuracy on Cityscapes by 1-2 points for strong dilated and skip architectures and by up to 10 points for suboptimal designs. Adapting receptive fields by dynamic Gaussian structure further improves results, equaling the accuracy of free-form deformation while improving efficiency.