IGCV3: Interleaved Low-Rank Group Convolutions for Efficient Deep Neural Networks

TL;DR

IGCV3 introduces a novel convolutional kernel combining structured sparse and low-rank patterns, leading to more efficient neural networks with improved performance on image classification and object detection tasks.

Contribution

This paper proposes a new convolutional kernel design that integrates sparse and low-rank structures, guided by a loose complementary condition, enhancing efficiency and accuracy.

Findings

Outperforms IGCV2 and MobileNetV2 on CIFAR and ImageNet

Achieves superior object detection results on COCO

Demonstrates improved efficiency and accuracy

Abstract

In this paper, we are interested in building lightweight and efficient convolutional neural networks. Inspired by the success of two design patterns, composition of structured sparse kernels, e.g., interleaved group convolutions (IGC), and composition of low-rank kernels, e.g., bottle-neck modules, we study the combination of such two design patterns, using the composition of structured sparse low-rank kernels, to form a convolutional kernel. Rather than introducing a complementary condition over channels, we introduce a loose complementary condition, which is formulated by imposing the complementary condition over super-channels, to guide the design for generating a dense convolutional kernel. The resulting network is called IGCV3. We empirically demonstrate that the combination of low-rank and sparse kernels boosts the performance and the superiority of our proposed approach to the state-of-the-arts, IGCV2 and MobileNetV2 over image classification on CIFAR and ImageNet and object detection on COCO.