Bootstrap Generalization Ability from Loss Landscape Perspective
This work addresses the problem of out-of-distribution generalization for computer vision practitioners, presenting an incremental approach by adapting existing loss landscape concepts to a new application.
The paper tackles domain generalization in computer vision by applying loss landscape theory to improve model generalization across unseen datasets, achieving third place in the ECCV 2022 NICO Challenge without domain invariant methods.
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.