Controllable Road Marking Generation
Controllable Road Marking Generation
This research introduces Controllable Road Marking Generation, a system that synthesizes road marking layouts from drivable-area masks and text descriptions. The method employs a text-conditioned latent rectified-flow DiT with topology-aware supervision and a Structured Gaussian Render post-process to recover crisp geometry. Evaluated on 4,597 Argoverse 2 test tiles, the system achieves a Buffered F1 of 76.8 and clDice of 50.2, outperforming an adapted mask-refinement baseline which scored 38.8 and 24.6 respectively. Independent training on the Waymo Open Dataset yielded scores of 88.0 for Buffered F1 and 66.2 for clDice.
The study presents a conditional bird's-eye-view pipeline for synthesizing road markings from masks and text, evaluated on Argoverse 2 and Waymo Open Dataset benchmarks with specific metric scores.
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.
Reported numbers
test tiles
4,597
Reported trials
View original evidence
n for thin, sparse markings, and (iii) Structured Gaussian Render (SGR), a training-free post-process that recovers crisp divider geometry by extracting polylines, fitting cubic B\'ezier curves, and re-rendering them as anisotropic super-Gaussian primitives. On 4,597 Argoverse~2 test tiles, our system achieves Buffered F1 of 76.8 and clDice of 50.2, compared with 38.8 and 24.6 for an adapted mask-refinement baseline.
Open source S4Argoverse 2 test tiles
4,597
Reported trials
View original evidence
n for thin, sparse markings, and (iii) Structured Gaussian Render (SGR), a training-free post-process that recovers crisp divider geometry by extracting polylines, fitting cubic B\'ezier curves, and re-rendering them as anisotropic super-Gaussian primitives. On 4,597 Argoverse~2 test tiles, our system achieves Buffered F1 of 76.8 and clDice of 50.2, compared with 38.8 and 24.6 for an adapted mask-refinement baseline.
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
n for thin, sparse markings, and (iii) Structured Gaussian Render (SGR), a training-free post-process that recovers crisp divider geometry by extracting polylines, fitting cubic B\'ezier curves, and re-rendering them as anisotropic super-Gaussian primitives. On 4,597 Argoverse~2 test tiles, our system achieves Buffered F1 of 76.8 and clDice of 50.2, compared with 38.8 and 24.6 for an adapted mask-refinement baseline.
Open source S4
When trained independently on the Waymo Open Dataset, it yields 88.0 Buffered F1 and 66.2 clDice. Component ablations show complementary connectivity gains from topology-aware supervision and SGR. Text-editing experiments reveal that stronger guidance improves edit success but also increases changes to non-target structures. We see this framework as a step toward simulation-ready road-marking variation, automated ma
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org