Unbiased Regression Loss for DETRs

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Edric, Daisuke, Ueta, Yukimasa, Kurokawa, Jayashree, Karlekar, Pranata, Sugiri
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909371794980864
author Edric
Daisuke, Ueta
Yukimasa, Kurokawa
Jayashree, Karlekar
Pranata, Sugiri
author_facet Edric
Daisuke, Ueta
Yukimasa, Kurokawa
Jayashree, Karlekar
Pranata, Sugiri
contents In this paper, we introduce a novel unbiased regression loss for DETR-based detectors. The conventional $L_{1}$ regression loss tends to bias towards larger boxes, as they disproportionately contribute more towards the overall loss compared to smaller boxes. Consequently, the detection performance for small objects suffers. To alleviate this bias, the proposed new unbiased loss, termed Sized $L_{1}$ loss, normalizes the size of all boxes based on their individual width and height. Our experiments demonstrate consistent improvements in both fully-supervised and semi-supervised settings using the MS-COCO benchmark dataset.
format Preprint
id arxiv_https___arxiv_org_abs_2410_22638
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Unbiased Regression Loss for DETRs
Edric
Daisuke, Ueta
Yukimasa, Kurokawa
Jayashree, Karlekar
Pranata, Sugiri
Computer Vision and Pattern Recognition
In this paper, we introduce a novel unbiased regression loss for DETR-based detectors. The conventional $L_{1}$ regression loss tends to bias towards larger boxes, as they disproportionately contribute more towards the overall loss compared to smaller boxes. Consequently, the detection performance for small objects suffers. To alleviate this bias, the proposed new unbiased loss, termed Sized $L_{1}$ loss, normalizes the size of all boxes based on their individual width and height. Our experiments demonstrate consistent improvements in both fully-supervised and semi-supervised settings using the MS-COCO benchmark dataset.
title Unbiased Regression Loss for DETRs
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2410.22638