CD-Depth: Unsupervised Domain Adaptation for Depth Estimation via Cross Domain Integration

Dongseok Shim, H. Jin Kim

Keywords: monocular depth estimation, unsupervised domain adaptation

Abstract: Despite the efficiency of data collecting for depth estimation in the synthetic environment, we cannot take full advantage of such benefit due to the distribution gap between the synthetic and the real world. In this paper, we introduce a new unsupervised domain adaptation framework, CD-Depth, for depth estimation to alleviate domain shift by extracting structure-consistent and domain-agnostic latents using following methods. (1) We propose domain-agnostic latent mapping which projects images from different domains to the shared latent space by removing redundant domain features for estimating monocular depth. (2) We also fuse visual signals from both RGB and latent domains to fully exploit multi domain information with adaptive-window-based cross-attention. Our proposed framework achieves state-of-the-art results in unsupervised domain adaptation for depth estimation both on indoor and outdoor datasets and produces better generalization performance on an unseen dataset.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics

Submission Guidelines: Yes

Please Choose The Closest Area That Your Submission Falls Into: Applications (eg, speech processing, computer vision, NLP)