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.
@article{arxiv.2209.08473,
title = {Bootstrap Generalization Ability from Loss Landscape Perspective},
author = {Huanran Chen and Shitong Shao and Ziyi Wang and Zirui Shang and Jin Chen and Xiaofeng Ji and Xinxiao Wu},
journal= {arXiv preprint arXiv:2209.08473},
year = {2023}
}
Comments
18 pages, 4 figures, Accepted by ECCV Workshop2022