Semi-Supervised Underwater Object Detection with Image Enhancement Guided by Attribute-based Data Distribution
Keywords: Semi-supervised learning; Underwater object detection
TL;DR: A novel underwater image enhancement method guided by attribute-based data distribution for semi-supervised underwater object detection
Abstract: Semi-supervised underwater object detection aims to improve the performance of detectors on unlabeled underwater images by leveraging knowledge from labeled ones. However, existing methods often overlook the distribution differences between labeled and unlabeled underwater images. In this paper, we propose a novel underwater image enhancement method guided by attribute-based data distribution (UIEG+), which focuses on reducing the discrepancies between enhanced and original unlabeled images across different attributes, thereby effectively addressing the challenges in semi-supervised underwater object detection. Specifically, we explore an underwater image enhancement strategy based on two attributes: color and scale distributions. For the color attribute, we construct a 3-dimensional grid memory, where each grid cell represents a color subspace and records the number of samples in that subspace. Similarly, for the scale attribute, we design a 1-dimensional vector memory that dynamically stores the number of samples in each scale subspace. Subsequently, we propose an effective sampling method to derive parameters for color and scale transformations based on the aforementioned distribution analysis, increasing the likelihood of transformations in low-distribution regions. To evaluate its effetiveness and superiority, massive semi-superivised underwater object deteciton experiments in multiple datasets have been conduted by integrating UIEG+ into existing semi-supervised object detection frameworks. The code will be released.
Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning
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: 2531
Loading