TY - GEN
T1 - Bootstrap Generalization Ability from Loss Landscape Perspective
AU - Chen, Huanran
AU - Shao, Shitong
AU - Wang, Ziyi
AU - Shang, Zirui
AU - Chen, Jin
AU - Ji, Xiaofeng
AU - Wu, Xinxiao
N1 - Publisher Copyright:
© 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.
PY - 2023
Y1 - 2023
N2 - Domain generalization aims to learn a model that can generalize well on the unseen test dataset, i.e., out-of-distribution data, which has different distribution from the training dataset. To address domain generalization in computer vision, we introduce the loss landscape theory into this field. Specifically, we bootstrap the generalization ability of the deep learning model from the loss landscape perspective in four aspects, including backbone, regularization, training paradigm, and learning rate. We verify the proposed theory on the NICO++, PACS, and VLCS datasets by doing extensive ablation studies as well as visualizations. In addition, we apply this theory in the ECCV 2022 NICO Challenge1 and achieve the 3rd place without using any domain invariant methods.
AB - Domain generalization aims to learn a model that can generalize well on the unseen test dataset, i.e., out-of-distribution data, which has different distribution from the training dataset. To address domain generalization in computer vision, we introduce the loss landscape theory into this field. Specifically, we bootstrap the generalization ability of the deep learning model from the loss landscape perspective in four aspects, including backbone, regularization, training paradigm, and learning rate. We verify the proposed theory on the NICO++, PACS, and VLCS datasets by doing extensive ablation studies as well as visualizations. In addition, we apply this theory in the ECCV 2022 NICO Challenge1 and achieve the 3rd place without using any domain invariant methods.
KW - Adaptive learning rate scheduler
KW - Distillation-based fine-tuning paradigm
KW - Domain generalization
KW - Loss landscape
KW - Wide-PyramidNet
UR - https://www.scopus.com/pages/publications/85150958017
U2 - 10.1007/978-3-031-25075-0_34
DO - 10.1007/978-3-031-25075-0_34
M3 - Conference contribution
AN - SCOPUS:85150958017
SN - 9783031250743
T3 - Lecture Notes in Computer Science
SP - 500
EP - 517
BT - Computer Vision – ECCV 2022 Workshops, Proceedings
A2 - Karlinsky, Leonid
A2 - Michaeli, Tomer
A2 - Nishino, Ko
PB - Springer Science and Business Media Deutschland GmbH
T2 - Workshops held at the 17th European Conference on Computer Vision, ECCV 2022
Y2 - 23 October 2022 through 27 October 2022
ER -