ACNet: Strengthening the Kernel Skeletons for Powerful CNN via Asymmetric Convolution Blocks

TL;DR

ACNet introduces Asymmetric Convolution Blocks (ACB) as a universal CNN component that enhances performance and robustness by strengthening kernel skeletons, applicable across various architectures without increasing inference costs.

Contribution

The paper proposes a novel architecture-neutral convolutional block, ACB, which improves CNN accuracy and robustness by replacing standard convolutions, then converting back to original architectures after training.

Findings

ACNet improves accuracy on CIFAR and ImageNet datasets.

ACB enhances robustness to rotational distortions.

ACNet requires no extra computation during inference.

Abstract

As designing appropriate Convolutional Neural Network (CNN) architecture in the context of a given application usually involves heavy human works or numerous GPU hours, the research community is soliciting the architecture-neutral CNN structures, which can be easily plugged into multiple mature architectures to improve the performance on our real-world applications. We propose Asymmetric Convolution Block (ACB), an architecture-neutral structure as a CNN building block, which uses 1D asymmetric convolutions to strengthen the square convolution kernels. For an off-the-shelf architecture, we replace the standard square-kernel convolutional layers with ACBs to construct an Asymmetric Convolutional Network (ACNet), which can be trained to reach a higher level of accuracy. After training, we equivalently convert the ACNet into the same original architecture, thus requiring no extra computations anymore. We have observed that ACNet can improve the performance of various models on CIFAR and ImageNet by a clear margin. Through further experiments, we attribute the effectiveness of ACB to its capability of enhancing the model's robustness to rotational distortions and strengthening the central skeleton parts of square convolution kernels.