RigidDriveGS: Object-Centric Vehicle Reconstruction from Fixed Multi-Camera Drive-Through Captures
RigidDriveGS: Object-Centric Vehicle Reconstruction from Fixed Multi-Camera Drive-Through Captures
This paper introduces RigidDriveGS, a pipeline for high-fidelity 3D reconstruction of vehicle exteriors using fixed multi-camera drive-through captures. The method handles challenges like rotating wheels and fisheye distortion through motion-gated tracking and mask-guided correspondences. Evaluated on 25 vehicles from dealership installations and phone recordings, the system achieves 30.42 dB PSNR and 0.90 SSIM, improving upon undistorted 3D-GS baselines without requiring controlled studio infrastructure.
The study evaluates RigidDriveGS on 25 vehicles captured from fixed multi-camera setups and handheld positions, reporting specific metrics like 30.42 dB PSNR and 0.026 m full body Chamfer-L1 against reference LiDAR scans.
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
dealership installations
10
View original evidence
centric pipeline combining motion-gated tracking, bifurcated masks that exclude wheels for SfM while optionally retaining them for rendering, mask-guided dense correspondences on raw distorted images, rig-constrained global SfM, and clean object-centric neural rendering with trimap-aware opacity or density supervision and native-alpha white compositing. We evaluate 25 vehicles from 10 dealership installations and six
Open source S3PSNR improvement
7.78
View original evidence
phone-recorded vehicles, each captured from 14 independent handheld positions. Under a staggered per-camera temporal-block holdout, 3DGUT+MCMC achieves 30.42 dB PSNR, 0.90 SSIM, and 0.20 LPIPS, improving PSNR by 7.78 dB over undistorted 3D-GS; independently recorded reference LiDAR scans yield full/body Chamfer-L1 of 0.026/0.013 m and F1@5 cm of 0.703/0.885. These results support clutter-free, object-centric novel-v
Open source S4Chamfer-L1 full body
0.026
View original evidence
phone-recorded vehicles, each captured from 14 independent handheld positions. Under a staggered per-camera temporal-block holdout, 3DGUT+MCMC achieves 30.42 dB PSNR, 0.90 SSIM, and 0.20 LPIPS, improving PSNR by 7.78 dB over undistorted 3D-GS; independently recorded reference LiDAR scans yield full/body Chamfer-L1 of 0.026/0.013 m and F1@5 cm of 0.703/0.885. These results support clutter-free, object-centric novel-v
Open source S4Chamfer-L1 body
0.013
View original evidence
phone-recorded vehicles, each captured from 14 independent handheld positions. Under a staggered per-camera temporal-block holdout, 3DGUT+MCMC achieves 30.42 dB PSNR, 0.90 SSIM, and 0.20 LPIPS, improving PSNR by 7.78 dB over undistorted 3D-GS; independently recorded reference LiDAR scans yield full/body Chamfer-L1 of 0.026/0.013 m and F1@5 cm of 0.703/0.885. These results support clutter-free, object-centric novel-v
Open source S4
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
centric pipeline combining motion-gated tracking, bifurcated masks that exclude wheels for SfM while optionally retaining them for rendering, mask-guided dense correspondences on raw distorted images, rig-constrained global SfM, and clean object-centric neural rendering with trimap-aware opacity or density supervision and native-alpha white compositing. We evaluate 25 vehicles from 10 dealership installations and six
Open source S3
phone-recorded vehicles, each captured from 14 independent handheld positions. Under a staggered per-camera temporal-block holdout, 3DGUT+MCMC achieves 30.42 dB PSNR, 0.90 SSIM, and 0.20 LPIPS, improving PSNR by 7.78 dB over undistorted 3D-GS; independently recorded reference LiDAR scans yield full/body Chamfer-L1 of 0.026/0.013 m and F1@5 cm of 0.703/0.885. These results support clutter-free, object-centric novel-v
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org