About training with ARKitScenes

[atomishcv]

Hi! I tried to train with the ARKitScenes Dataset, you said, "We mainly conduct experiments on the HyperSim synthetic dataset [49] and two real-world datasets: ScanNet++ [70] and ARKitScenes [3], which provide iPhone RGB-LiDAR data (192 × 256 resolution) and annotated depth from a high-power LiDAR (1440 × 1920 resolution)."
But the high-power depth has noise. How did you solve this problem? use Zip-Nerf?
>