TOSN-Trans：Transparent Object Segmentation Network with Transformer
Keywords: RGB-D, glass segmentation, feature fusion, dual-network, efficient transformer, boundary optimization
Abstract: Due to the optical properties of glass materials, most glass appears transparent in RGB images. However, in depth images, different acquisition methods make glass visible. Therefore, Therefore, using RGB-D dual-channel feature input makes it easier to recognize and segment glass objects. Building on this concept, we propose a multi-layer symmetrical dual-channel network architecture, which can effectively realize trans-modal feature fusion of RGB-D images based on attention mechanism, and integrate Convolutional and Transformer architectures to extract local features and non-local dependencies, respectively. To further enhance segmentation accuracy and efficiency, this paper also designs a boundary optimization module. This module constructs a distance map based on edge prediction guidance, enabling high-precision glass edge recognition. To support this work, we collect a new dataset comprising 5551 sets of calibrated RGB-D images. The effectiveness and accuracy of the proposed glass segmentation method are rigorously evaluated quantitatively and qualitatively. The code for this paper has been published at:https://github.com/Jaccury/RGB-D-Transparent-object-segmentation.
Primary Area: applications to computer vision, audio, language, and other modalities
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: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
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.
Submission Number: 5761
Loading