Improving the Robustness of Deep Neural Networks via Stability Training

TL;DR

This paper introduces a stability training method to enhance deep neural networks' robustness against input perturbations like compression and cropping, improving performance on various vision tasks.

Contribution

A novel stability training approach that significantly improves neural network robustness to input distortions across multiple computer vision applications.

Findings

Enhanced robustness of Inception model against image distortions

Improved performance on near-duplicate detection and noisy datasets

Achieved state-of-the-art results in robustness benchmarks

Abstract

In this paper we address the issue of output instability of deep neural networks: small perturbations in the visual input can significantly distort the feature embeddings and output of a neural network. Such instability affects many deep architectures with state-of-the-art performance on a wide range of computer vision tasks. We present a general stability training method to stabilize deep networks against small input distortions that result from various types of common image processing, such as compression, rescaling, and cropping. We validate our method by stabilizing the state-of-the-art Inception architecture against these types of distortions. In addition, we demonstrate that our stabilized model gives robust state-of-the-art performance on large-scale near-duplicate detection, similar-image ranking, and classification on noisy datasets.