Optimizing Grouped Convolutions on Edge Devices

TL;DR

This paper introduces Grouped Spatial Pack Convolutions (GSPC), a novel implementation that significantly accelerates grouped convolutions on edge devices, outperforming existing solutions in speed and scalability.

Contribution

We propose GSPC, a new grouped convolution implementation optimized for edge devices, and demonstrate its superior performance over existing methods in popular frameworks.

Findings

GSPC outperforms existing implementations by up to 8x in inference speed.

The implementation scales well with the number of groups.

GSPC achieves consistent speed improvements across multiple edge devices.

Abstract

When deploying a deep neural network on constrained hardware, it is possible to replace the network's standard convolutions with grouped convolutions. This allows for substantial memory savings with minimal loss of accuracy. However, current implementations of grouped convolutions in modern deep learning frameworks are far from performing optimally in terms of speed. In this paper we propose Grouped Spatial Pack Convolutions (GSPC), a new implementation of grouped convolutions that outperforms existing solutions. We implement GSPC in TVM, which provides state-of-the-art performance on edge devices. We analyze a set of networks utilizing different types of grouped convolutions and evaluate their performance in terms of inference time on several edge devices. We observe that our new implementation scales well with the number of groups and provides the best inference times in all settings, improving the existing implementations of grouped convolutions in TVM, PyTorch and TensorFlow Lite by 3.4x, 8x and 4x on average respectively. Code is available at https://github.com/gecLAB/tvm-GSPC/