Geometric Refinement and Distributional Alignment for Sim2Real Roadside LiDAR Detection
Geometric Refinement and Distributional Alignment for Sim2Real Roadside LiDAR Detection
This paper presents an approach for the LUMPI track of the UrbanTwin Sim2Real LiDAR Challenge, achieving a combined score of 0.4793 using a PointPillars baseline trained on synthetic data and evaluated on real roadside LiDAR. Key contributions include a parameter-free geometric refinement improving pedestrian AP by 86%, a post-processing repair boosting Van class detection by 335%, and an analytical method for realism metrics that allowed the authors to predict their final score to four decimal places before submission.
The authors attribute their performance gains to geometric refinement, post-processing repair, and analytical treatment of realism metrics rather than architectural changes.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Data origin
- Synthetic dataOpen source S2
Reported numbers
realism score
0.9126
View original evidence
a realism score of 0.9126. Our contribution is not a new architecture. Working from a PointPillars baseline, we obtained most of our gains from three sources. First, a diagnosis that small object failure in roadside LiDAR is a localisation problem rather than a detection problem, which led to a parameter free geometric refinement worth 86 percent relative on pedestrian AP. Second, an audit of our own post processing
Open source S3detection mAP
0.1333
View original evidence
arXiv:2611.03936v1 Announce Type: cross Abstract: We describe our approach to the LUMPI track of the UrbanTwin Sim2Real LiDAR Challenge, in which a 3D object detector must be trained exclusively on synthetic data and evaluated on unseen real roadside LiDAR, while the synthetic point clouds are simultaneously scored for distributional realism. Our final entry scores 0.4793 combined, from a detection mAP of 0.1333 and
Open source S2combined score
0.4793
View original evidence
arXiv:2611.03936v1 Announce Type: cross Abstract: We describe our approach to the LUMPI track of the UrbanTwin Sim2Real LiDAR Challenge, in which a 3D object detector must be trained exclusively on synthetic data and evaluated on unseen real roadside LiDAR, while the synthetic point clouds are simultaneously scored for distributional realism. Our final entry scores 0.4793 combined, from a detection mAP of 0.1333 and
Open source S2
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
arXiv:2611.03936v1 Announce Type: cross Abstract: We describe our approach to the LUMPI track of the UrbanTwin Sim2Real LiDAR Challenge, in which a 3D object detector must be trained exclusively on synthetic data and evaluated on unseen real roadside LiDAR, while the synthetic point clouds are simultaneously scored for distributional realism. Our final entry scores 0.4793 combined, from a detection mAP of 0.1333 and
Open source S2
a realism score of 0.9126. Our contribution is not a new architecture. Working from a PointPillars baseline, we obtained most of our gains from three sources. First, a diagnosis that small object failure in roadside LiDAR is a localisation problem rather than a detection problem, which led to a parameter free geometric refinement worth 86 percent relative on pedestrian AP. Second, an audit of our own post processing
Open source S3
transferable result. Because mean AP decomposes into independent per class terms, a single submission can carry several mutually controlled experiments, and a submission assembled entirely from previously measured components has a score that is known before it is sent. We predicted our final score to four decimal places prior to submission. We report negative results at equal length, and give a full account of data
Open source S5
Implications for data suppliers
RoboSignal interpretation and collection questions, not statements of buyer demand.
- Ask which physical-world evaluation and real-data validation the buyer requires for simulated data.
- Compare the reported units and scope before using these quantities in a budget. Recording hours, sensor-hours and trajectories are different measures.
- Validate demand and acceptance criteria with a buyer before scaling. Publication, popularity and a research result do not establish a purchase commitment.
Source：arXiv Robotics — research abstracts · arxiv.org