A step towards understanding why classification helps regression

S. Pintea, Yancong Lin, J. Dijkstra, J. V. Gemert
8/1/2023

No official code found. Best implementation: SilviaLauraPintea/reg-cls (6 stars, last commit 2023-12-27).

Abstract

A number of computer vision deep regression approaches report improved results when adding a classification loss to the regression loss. Here, we explore why this is useful in practice and when it is beneficial. To do so, we start from precisely controlled dataset variations and data samplings and find that the effect of adding a classification loss is the most pronounced for regression with imbalanced data. We explain these empirical findings by formalizing the relation between the balanced and imbalanced regression losses. Finally, we show that our findings hold on two real imbalanced image datasets for depth estimation (NYUD2-DIR), and age estimation (IMDB-WIKI-DIR), and on the problem of imbalanced video progress prediction (Breakfast). Our main takeaway is: for a regression task, if the data sampling is imbalanced, then add a classification loss.

View on arXivView PDF

Code Implementations(1)

SilviaLauraPintea/reg-clsCommunity100%

Python implementation of the experiments from "A step towards understanding why classification helps regression", Pintea, Silvia L and Yancong, Lin and Dijkstra, Jouke and van Gemert, Jan C, ICCV, 2023.

61Shell, PythonOct 9, 20232 years ago

Similar Papers

Imbalance in Regression Datasets

83% match2024

Deep Imbalanced Regression via Hierarchical Classification Adjustment

83% match1 repo52023

Dist Loss: Enhancing Regression in Few-Shot Region through Distribution Distance Constraint

81% match1 repo102024

Improving Deep Regression with Ordinal Entropy

81% match1 repo542023

Enlarged Large Margin Loss for Imbalanced Classification

80% match1 repo42023

Towards Understanding How Data Augmentation Works with Imbalanced Data

80% match1 repo2023

Cite this paper

@article{pintea2023step,
  title  = {A step towards understanding why classification helps regression},
  author = {S. Pintea and Yancong Lin and J. Dijkstra and J. V. Gemert},
  year   = {2023},
  eprint = {2308.10603},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/pdf/2308.10603},
  journal = {ICCV 2023 2023}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2308.10603.svg)](https://codeofpaper.com/paper/2308.10603)

Discussion