A Comprehensive Analysis of Deep Regression

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.

arXiv commentPublished in IEEE TPAMI

References in corpus (16)

• Adam: A Method for Stochastic Optimization
• Very Deep Convolutional Networks for Large-Scale Image Recognition
• Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift
• LSTM: A Search Space Odyssey
• ADADELTA: An Adaptive Learning Rate Method
• How transferable are features in deep neural networks?
• DeepPose: Human Pose Estimation via Deep Neural Networks
• Convolutional Neural Networks for Medical Image Analysis: Full Training or Fine Tuning?
• How far are we from solving the 2D & 3D Face Alignment problem? (and a dataset of 230,000 3D facial landmarks)
• Joint Training of a Convolutional Network and a Graphical Model for Human Pose Estimation
• Systematic evaluation of CNN advances on the ImageNet
• Convolutional Neural Fabrics
• P-values: misunderstood and misused
• Robust Head-Pose Estimation Based on Partially-Latent Mixture of Linear Regressions
• Deep Convolutional Neural Network Design Patterns
• On architectural choices in deep learning: From network structure to gradient convergence and parameter estimation

Cited by in corpus (32)

• A Review on Deep Learning in UAV Remote Sensing
• Data driven approximation of parametrized PDEs by Reduced Basis and Neural Networks
• From WiscKey to Bourbon: A Learned Index for Log-Structured Merge Trees
• Densely connected neural networks for nonlinear regression
• CAHPHF: Context-Aware Hierarchical QoS Prediction with Hybrid Filtering
• How to Train Your Energy-Based Model for Regression
• DPNNet-2.0 Part I: Finding hidden planets from simulated images of protoplanetary disk gaps
• Consistent and Flexible Selectivity Estimation for High-Dimensional Data
• Deep Learning in Wide-field Surveys: Fast Analysis of Strong Lenses in Ground-based Cosmic Experiments
• RSSI-Based Hybrid Beamforming Design with Deep Learning
• Forklens: Accurate weak-lensing shear measurement with deep learning
• Optimal STEM Convergence Angle Selection using a Convolutional Neural Network and the Strehl Ratio
• On Dropout and Nuclear Norm Regularization
• Variational Tensor Neural Networks for Deep Learning
• A Neural Anthropometer Learning from Body Dimensions Computed on Human 3D Meshes
• Generative models-based data labeling for deep networks regression: application to seed maturity estimation from UAV multispectral images
• Backdoor Attack and Defense for Deep Regression
• DeepGUM: Learning Deep Robust Regression with a Gaussian-Uniform Mixture Model
• End-to-End Learning of Speech 2D Feature-Trajectory for Prosthetic Hands
• Sparse bottleneck neural networks for exploratory non-linear visualization of Patch-seq data
• Multi-Sensor Diffusion-Driven Optical Image Translation for Large-Scale Applications
• Survey of Deep Learning Methods for Inverse Problems
• On the benefits of maximum likelihood estimation for Regression and Forecasting
• Robust and Active Learning for Deep Neural Network Regression
• Overestimation learning with guarantees
• Neural Scaling Laws for Deep Regression
• DeepRed: an architecture for redshift estimation
• Deep Multimodal Transfer-Learned Regression in Data-Poor Domains
• FES: A Fast Efficient Scalable QoS Prediction Framework
• Learning Visual Voice Activity Detection with an Automatically Annotated Dataset
• Complete CVDL Methodology for Investigating Hydrodynamic Instabilities
• Using Robust Regression to Find Font Usage Trends