Submitted by JUNHONG MIN.

Submission data

Full name	Confidence-Aware Symmetric Stereo Matching via U-Net Transformer
Description	we propose a novel deep stereo matching network a new real-world stereo dataset of cluttered objects taken with a commercially available stereo sensor. We design a U-shaped architecture with various types of attentions which more efficiently extracts global and local contexts from rectified image pairs, resulting in highly accurate disparities. Furthermore, its symmetric structure allows simultaneous estimation both left and right disparity. It can also implicitly estimate the uncertainty i.e. the confidence of estimated disparities.
Parameters	feature dimension 128: 4-level U-net for the feature extraction 3-level U-net for the refinement processing at 1536x1024 pixel scale
Publication title	Confidence-Aware Symmetric Stereo Matching via U-Net Transformer
Publication authors	Junhong Min, Youngpil Jeon
Programming language(s)	Python
Hardware	i9-13900k, Nvidia RTX 4090
Submission creation date	14 Sep, 2023
Last edited	14 Sep, 2023

High-res multi-view results

Low-res many-view results

Low-res two-view results

SLAM results