SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2308.10603 · ICCV · 2023

A step towards understanding why classification helps regression

Jan Van Gemert, Silvia Pintea, Yancong Lin, Jouke Dijkstra

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 14 functions out of this paper's own repositories and ran 14 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

RepositoryRoleRan
silvialaurapintea/reg-cls canonical 14 of 14
FunctionStatusWhere it lives
accuracy Ran silvialaurapintea/reg-cls/1d/utils.py
pointer only (licence: NONE) · get_code("17fb0539d6b732dd")
change_ranges Ran silvialaurapintea/reg-cls/1d/histeq.py
pointer only (licence: NONE) · get_code("21fe9de526aff373")
conv3x3 Ran silvialaurapintea/reg-cls/images/imdb-wiki-dir/resnet.py
pointer only (licence: NONE) · get_code("fac5364e2f53c6db")
def_dataloaders Ran silvialaurapintea/reg-cls/1d/train_eval.py
pointer only (licence: NONE) · get_code("2737bfb410c85be7")
doEq Ran silvialaurapintea/reg-cls/1d/histeq.py
pointer only (licence: NONE) · get_code("713b88d78142fdf6")
func Ran silvialaurapintea/reg-cls/1d/make_data1d.py
pointer only (licence: NONE) · get_code("af188df6229ac760")
get_imbalanced Ran silvialaurapintea/reg-cls/1d/make_data1d.py
pointer only (licence: NONE) · get_code("d149be207b41dae6")
get_uniform Ran silvialaurapintea/reg-cls/1d/make_data1d.py
pointer only (licence: NONE) · get_code("c6c46e08e8618f36")
removeStartZeros Ran silvialaurapintea/reg-cls/1d/histeq.py
pointer only (licence: NONE) · get_code("6e97647f8b72f5a8")
resume Ran silvialaurapintea/reg-cls/1d/train_eval.py
pointer only (licence: NONE) · get_code("9d857dee26c989e2")
shot_metrics Ran silvialaurapintea/reg-cls/images/imdb-wiki-dir/train_cls.py
pointer only (licence: NONE) · get_code("6e71bc6cd8ccb243")
weighted_focal_mse_loss Ran silvialaurapintea/reg-cls/images/imdb-wiki-dir/loss.py
pointer only (licence: NONE) · get_code("953f661a54a1b2bd")
weighted_l1_loss Ran silvialaurapintea/reg-cls/images/imdb-wiki-dir/loss.py
pointer only (licence: NONE) · get_code("2a25dd59364377ff")
weighted_mse_loss Ran silvialaurapintea/reg-cls/images/imdb-wiki-dir/loss.py
pointer only (licence: NONE) · get_code("1da83cb2ed411cd0")

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

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.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2308.10603")
get_code_for_paper("2308.10603")
have("2308.10603")

Connect an agent — have() is free.