Pose optimization for autonomous driving datasets using neural rendering models

arXiv 2025

Novel View Synthesis Benchmark

  • We evaluate Novel View Synthesis methods using our optimized poses and the original poses.
  • For the metrics, we evaluate the PSNR, SSIM and LPIPS metrics after training with 1 out of 2 frames and the default parameters.
  • The results for each dataset are the average values over 5 selected sequences.
  • We name our poses HW, which are optimized through the process of the paper.

KITTI-360

Waymo

NuScenes

PandaSet