Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning

TL;DR

This paper investigates the integration of residual connections into Inception architectures, demonstrating that residuals accelerate training, slightly improve accuracy, and enable the development of more effective deep convolutional networks for image recognition.

Contribution

It provides empirical evidence that residual connections benefit Inception networks by speeding up training and marginally improving performance, along with new streamlined architectures for better recognition.

Findings

Residual connections significantly accelerate Inception training.

Residual Inception networks slightly outperform non-residual counterparts.

Ensemble of residual and Inception-v4 achieves 3.08% top-5 error on ImageNet.

Abstract

Very deep convolutional networks have been central to the largest advances in image recognition performance in recent years. One example is the Inception architecture that has been shown to achieve very good performance at relatively low computational cost. Recently, the introduction of residual connections in conjunction with a more traditional architecture has yielded state-of-the-art performance in the 2015 ILSVRC challenge; its performance was similar to the latest generation Inception-v3 network. This raises the question of whether there are any benefit in combining the Inception architecture with residual connections. Here we give clear empirical evidence that training with residual connections accelerates the training of Inception networks significantly. There is also some evidence of residual Inception networks outperforming similarly expensive Inception networks without residual connections by a thin margin. We also present several new streamlined architectures for both residual and non-residual Inception networks. These variations improve the single-frame recognition performance on the ILSVRC 2012 classification task significantly. We further demonstrate how proper activation scaling stabilizes the training of very wide residual Inception networks. With an ensemble of three residual and one Inception-v4, we achieve 3.08 percent top-5 error on the test set of the ImageNet classification (CLS) challenge