Adaptive aggregation and dynamic self-learning for multi-view stereo network

Xiaoyan Zhang; Xiang Chen; Hao Shi; Pan Luo

doi:10.1117/12.2653490

8 December 2022 Adaptive aggregation and dynamic self-learning for multi-view stereo network

Xiaoyan Zhang, Xiang Chen, Hao Shi, Pan Luo

Proceedings Volume 12474, Second International Symposium on Computer Technology and Information Science (ISCTIS 2022); 124742J (2022) https://doi.org/10.1117/12.2653490
Event: Second International Symposium on Computer Technology and Information Science (ISCTIS 2022), 2022, Guilin, China

Abstract

For the problem of poor accuracy and integrity of multi-view stereo (MVS) reconstruction, we propose an efficient multi-view stereo network, which mainly studies feature extraction module and depth map optimization module. At first, we introduce adaptive aggregation module to use context-aware convolution and multi-scale aggregation to adaptively extract image features, which effectively improves the extraction accuracy of weak texture surface. Then, we introduce dynamic self-learning optimization module (DSL) to solve the problem of depth map over-smoothing. Experiments show that, compared with MVSNet, our network greatly improves the integrity of reconstruction, and does not increase memory and time overhead.

Citation Download Citation

Xiaoyan Zhang, Xiang Chen, Hao Shi, and Pan Luo "Adaptive aggregation and dynamic self-learning for multi-view stereo network", Proc. SPIE 12474, Second International Symposium on Computer Technology and Information Science (ISCTIS 2022), 124742J (8 December 2022); https://doi.org/10.1117/12.2653490

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available