期刊
2022 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2022)
卷 -, 期 -, 页码 16907-16916出版社
IEEE COMPUTER SOC
DOI: 10.1109/CVPR52688.2022.01642
关键词
-
资金
- Samsung Electronics Co., Ltd [G01200447]
- National Research Foundation of Korea [NRF-2020M3H8A1115028]
- National Research Foundation of Korea [2020M3H8A1115028] Funding Source: Korea Institute of Science & Technology Information (KISTI), National Science & Technology Information Service (NTIS)
This paper proposes a new multi-modal test-time adaptation approach for 3D semantic segmentation. It includes two modules, intra-modal pseudo-label generation and inter-modal pseudo-label refinement, which provide stable self-learning signals in various multi-modal test-time adaptation scenarios.
Test-time adaptation approaches have recently emerged as a practical solution for handling domain shift without access to the source domain data. In this paper, we propose and explore a new multi-modal extension of test-time adaptation for 3D semantic segmentation. We find that, directly applying existing methods usually results in performance instability at test time, because multi-modal input is not considered jointly. To design a framework that can take All advantage of multi-modality, where each modality provides regularized self-supervisory signals to other modalities, we propose two complementary modules within and across the modalities. First, Intra-modal Pseudo-label Generation (Infra-PG) is introduced to obtain reliable pseudo labels within each modality by aggregating information from two models that are both pre-trained on source data but updated with target data at different paces. Second, Inter-modal Pseudo-label Refinement (Inter-PR) adaptively selects more reliable pseudo labels from different modalities based on a proposed consistency scheme. Experiments demonstrate that our regularized pseudo labels produce stable self-learning signals in numerous multi-modal test-time adaptation scenarios for 3D semantic segmentation. Visit our project website at https: //www.nec-labs.com/mas/MM-TTA
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据