Multi-Task Perception in Unstructured Environments: Anti-Degradation Complementary Learning and SAMEnhancer
Keywords: autonomous driving, unstructured environment, mobile sam, semi-supervised learning
Abstract: While autonomous driving perception has advanced significantly in structured environments, unstructured environments still present major challenges due to the complexity of traffic participants and irregular road conditions. This paper focuses on addressing these challenges through multi-task perception, targeting drivable area segmentation and object detection in unstructured environments. A key issue in existing datasets for unstructured settings is the non-overlapping annotation of images across different tasks, which limits the efficiency of data utilization. To tackle this, we propose Anti-Degradation Complementary Learning (ADC learning), a semi-supervised approach that allows different tasks to share knowledge across unlabeled data, thereby maximizing the use of available image information. Additionally, we introduce SAMEnhancer, which integrates the Segment Anything Model (SAM) to improve segmentation quality by combining the semantic specificity of network training with the coherence of SAM’s segmentation. Extensive experiments validate the effectiveness of our methods, demonstrating significant performance improvements in both segmentation and detection, especially in challenging unstructured scenarios.
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: 9710
Loading