MindLoc: A Secure Brain-Based System for Object Localization
Keywords: Multimodal, Privacy Protection, fMRI, Object Localization
TL;DR: We introduce MindLoc, a cross-subject brain-based object localization model that combines precision and convenience, utilizing fMRI signals and encryption technology to enhance privacy and performance, achieving state-of-the-art results.
Abstract: Object localization tasks aim to accurately locate and identify specified target objects within images, representing a core challenge in the field of computer vision. Traditional object localization systems primarily rely on intermediary modalities such as text descriptions, speech, or visual cues to interpret human intent. However, these modalities only provide indirect expressions of human intent, limiting the efficiency of information transmission. This is particularly evident when detailed descriptions of texture and spatial information are required, resulting in higher interaction costs. While existing brain-based object localization systems offer the potential for directly interpreting human intent, their localization accuracy still lags behind traditional text-based systems. Additionally, the high cost of data collection, limited diversity of participants, and significant individual cognitive differences make it challenging to train subject-independent models, thereby constraining the development of brain-based object localization systems. To address the challenges, we propose MindLoc, a lightweight, cross-subject brain-based object localization model. MindLoc can rapidly and accurately locate target objects in complex images by directly analyzing fMRI signals, combining the precision of traditional localization systems with the convenience of brain-based systems. Additionally, we are the first to introduce encryption technology for the privacy protection of brain data, significantly reducing the psychological burden on participants, which provides a foundation for increasing participant diversity in future studies. Experimental results demonstrate that MindLoc has achieved new state-of-the-art performance in brain-based object localization tasks, showcasing significant advantages in both accuracy and convenience. Our code is
available at https://mindloc-sys.github.io/.
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: 4692
Loading