A Comprehensive Analysis of Deep Regression

TL;DR

This paper systematically evaluates deep regression models in vision tasks, revealing that network architecture changes have less impact than data pre-processing, and that general-purpose networks can achieve near state-of-the-art results.

Contribution

First comprehensive analysis of deep regression techniques, including statistical evaluation across multiple vision tasks and insights into the impact of data pre-processing.

Findings

Data pre-processing variability exceeds architecture variability.

General-purpose networks like VGG-16 and ResNet-50 perform near state-of-the-art.

Statistical confidence intervals provided for performance metrics.

Abstract

Deep learning revolutionized data science, and recently its popularity has grown exponentially, as did the amount of papers employing deep networks. Vision tasks, such as human pose estimation, did not escape from this trend. There is a large number of deep models, where small changes in the network architecture, or in the data pre-processing, together with the stochastic nature of the optimization procedures, produce notably different results, making extremely difficult to sift methods that significantly outperform others. This situation motivates the current study, in which we perform a systematic evaluation and statistical analysis of vanilla deep regression, i.e. convolutional neural networks with a linear regression top layer. This is the first comprehensive analysis of deep regression techniques. We perform experiments on four vision problems, and report confidence intervals for the median performance as well as the statistical significance of the results, if any. Surprisingly, the variability due to different data pre-processing procedures generally eclipses the variability due to modifications in the network architecture. Our results reinforce the hypothesis according to which, in general, a general-purpose network (e.g. VGG-16 or ResNet-50) adequately tuned can yield results close to the state-of-the-art without having to resort to more complex and ad-hoc regression models.