Affordable Access

deepdyve-link
Publisher Website

A Comprehensive Analysis of Deep Regression.

Authors
  • Lathuiliere, Stephane
  • Mesejo, Pablo
  • Alameda-Pineda, Xavier
  • Horaud, Radu
Type
Published Article
Journal
IEEE transactions on pattern analysis and machine intelligence
Publication Date
Sep 01, 2020
Volume
42
Issue
9
Pages
2065–2081
Identifiers
DOI: 10.1109/TPAMI.2019.2910523
PMID: 30990175
Source
Medline
Language
English
License
Unknown

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.

Report this publication

Statistics

Seen <100 times