3D-PL: Domain Adaptive Depth Estimation with 3D-Aware Pseudo-Labeling

Yu Ting Yen, Chia Ni Lu, Wei Chen Chiu*, Yi Hsuan Tsai


研究成果: Conference contribution同行評審


For monocular depth estimation, acquiring ground truths for real data is not easy, and thus domain adaptation methods are commonly adopted using the supervised synthetic data. However, this may still incur a large domain gap due to the lack of supervision from the real data. In this paper, we develop a domain adaptation framework via generating reliable pseudo ground truths of depth from real data to provide direct supervisions. Specifically, we propose two mechanisms for pseudo-labeling: 1) 2D-based pseudo-labels via measuring the consistency of depth predictions when images are with the same content but different styles; 2) 3D-aware pseudo-labels via a point cloud completion network that learns to complete the depth values in the 3D space, thus providing more structural information in a scene to refine and generate more reliable pseudo-labels. In experiments, we show that our pseudo-labeling methods improve depth estimation in various settings, including the usage of stereo pairs during training. Furthermore, the proposed method performs favorably against several state-of-the-art unsupervised domain adaptation approaches in real-world datasets.

主出版物標題Computer Vision – ECCV 2022 - 17th European Conference, 2022, Proceedings
編輯Shai Avidan, Gabriel Brostow, Moustapha Cissé, Giovanni Maria Farinella, Tal Hassner
發行者Springer Science and Business Media Deutschland GmbH
出版狀態Published - 2022
事件17th European Conference on Computer Vision, ECCV 2022 - Tel Aviv, Israel
持續時間: 23 10月 202227 10月 2022


名字Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
13687 LNCS


Conference17th European Conference on Computer Vision, ECCV 2022
城市Tel Aviv


深入研究「3D-PL: Domain Adaptive Depth Estimation with 3D-Aware Pseudo-Labeling」主題。共同形成了獨特的指紋。