ACLS: Adaptive and Conditional Label Smoothing for Network Calibration

Authors
Park, HyekangNoh, JongyounOh, YoungminBaek, DonghyeonHam, Bumsub
Issue Date
2023-10
Publisher
IEEE COMPUTER SOC
Citation
IEEE/CVF International Conference on Computer Vision (ICCV), pp.3913 - 3922
Abstract
We address the problem of network calibration adjusting miscalibrated confidences of deep neural networks. Many approaches to network calibration adopt a regularization-based method that exploits a regularization term to smooth the miscalibrated confidences. Although these approaches have shown the effectiveness on calibrating the networks, there is still a lack of understanding on the underlying principles of regularization in terms of network calibration. We present in this paper an in-depth analysis of existing regularization-based methods, providing a better understanding on how they affect to network calibration. Specifically, we have observed that 1) the regularization-based methods can be interpreted as variants of label smoothing, and 2) they do not always behave desirably. Based on the analysis, we introduce a novel loss function, dubbed ACLS, that unifies the merits of existing regularization methods, while avoiding the limitations. We show extensive experimental results for image classification and semantic segmentation on standard benchmarks, including CIFAR10, Tiny-ImageNet, ImageNet, and PASCAL VOC, demonstrating the effectiveness of our loss function.
ISSN
1550-5499
URI
https://pubs.kist.re.kr/handle/201004/149670
DOI
10.1109/ICCV51070.2023.00364
Appears in Collections:
KIST Conference Paper > 2023
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML

qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE