Convolution with even-sized kernels and symmetric padding

TL;DR

This paper introduces symmetric padding for even-sized convolution kernels, addressing the shift problem and enabling these kernels to outperform traditional 3x3 kernels in image tasks with less computational cost.

Contribution

Proposes symmetric padding techniques (C2sp, C4sp) to eliminate the shift problem in even-sized kernels, improving their effectiveness and efficiency in neural network architectures.

Findings

Symmetric padding alleviates the shift problem in even-sized kernels.

Even-sized kernels with symmetric padding outperform 3x3 kernels in classification and generation tasks.

C2sp achieves comparable accuracy to compact models with less training memory and time.

Abstract

Compact convolutional neural networks gain efficiency mainly through depthwise convolutions, expanded channels and complex topologies, which contrarily aggravate the training process. Besides, 3x3 kernels dominate the spatial representation in these models, whereas even-sized kernels (2x2, 4x4) are rarely adopted. In this work, we quantify the shift problem occurs in even-sized kernel convolutions by an information erosion hypothesis, and eliminate it by proposing symmetric padding on four sides of the feature maps (C2sp, C4sp). Symmetric padding releases the generalization capabilities of even-sized kernels at little computational cost, making them outperform 3x3 kernels in image classification and generation tasks. Moreover, C2sp obtains comparable accuracy to emerging compact models with much less memory and time consumption during training. Symmetric padding coupled with even-sized convolutions can be neatly implemented into existing frameworks, providing effective elements for architecture designs, especially on online and continual learning occasions where training efforts are emphasized.