Unsupervised Radar Point Cloud Enhancement Using Diffusion Model as Prior without Paired Traning Data
Keywords: Radar, Point Cloud Enhancement, Diffusion Model, Inverse Problem, Autonomous Driving
TL;DR: This paper proposes a method that construct radar point cloud enhancement as an inverse problem which can be improved by using a diffusion model.
Abstract: In industrial automation technology, radar is one of the crucial sensors in the machine perception stage. However, due to the long wavelength of radar electromagnetic waves and the limited number of antennas, the angle resolution is limited. Recent advancements have introduced methods that leverage paired LiDAR-radar data for training, achieving notable point enhancement effect. However, the requirement for paired data significantly increases the cost and complexity of model development, limiting model’s widespread adoption and scalability. To address this, we propose an unsupervised radar point cloud enhancement algorithm using diffusion model as prior without paired training data. Specifically, our method formulates radar angle estimation recovery into an inverse problem and introduces prior knowledge via a diffusion model when solving it. Experimental results demonstrate that our method achieves high fidelity and low noise performance compared to traditional regularization methods. Compared to paired data training methods, our approach not only delivers comparable performance but also offers greater content control and reduced generation variance. Additionally, it does not require a huge amount of paired data. To the best of our knowledge, our method is the first to enhance radar point cloud by introducing prior knowledge via diffusion model instead of training on paired data.
Primary Area: applications to robotics, autonomy, planning
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: 10629
Loading