FDTDNet: Privacy-Preserving Lensless Object Segmentation via Feature Demultiplexing and Task Decoupling
Keywords: Lensless Object Segmentation; Lensless Imaging; Privacy-Preserving; Feature Demultiplexing; Task Decoupling
Abstract: Camera-based vision systems pose privacy risks, whereas lensless cameras present a viable alternative by omitting visual semantics from their measurements due to the absence of lenses. However, these captured lensless measurements pose challenges for existing computer vision tasks such as object segmentation that usually require visual input. To address this problem, we propose a lensless object segmentation network via feature demultiplexing and task decoupling (FDTDNet) to perform object segmentation for lensless measurements. Specifically, we propose an optical-aware feature demultiplexing mechanism to get meaningful features from lensless measurements without visual reconstruction and design a multi-task learning framework decoupling the lensless object segmentation task into two subtasks, i.e., the reason for contour distribution maps (CDM) and body distribution maps (BDM), respectively. Extensive experiments demonstrate that our FDTDNet achieves highly accurate segmentation effect, which sheds light on privacy-preserving high-level vision with compact lensless cameras.
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.
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: 6278
Loading