Skip to content

Pre-trained weights

Latest
Compare
Choose a tag to compare
@kazuto1011 kazuto1011 released this 26 Oct 03:04
· 19 commits to main since this release
0f7c2da

Usage:

import torch

ddpm, lidar_utils, cfg = torch.hub.load(
    repo_or_dir = "kazuto1011/r2dm",
    model       = "pretrained_r2dm",
    config      = "r2dm-h-kitti360-300k", # default
    device      = "cuda",
)

KITTI-360 (64x1024, spherical projection)

Config Loss Depth Positional Encoding FRD (T=256) ↓ torch.load kwargs
A L2 Log Identity 202.40 config="r2dm-a-kitti360-300k"
B L1 Log Identity 382.35 config="r2dm-b-kitti360-300k"
C Huber Log Identity 174.83 config="r2dm-c-kitti360-300k"
D L2 Metric Identity 229.28 config="r2dm-d-kitti360-300k"
E L2 Inverse Identity 188.84 config="r2dm-e-kitti360-300k"
F L2 Log w/o spatial bias 910.67 config="r2dm-f-kitti360-300k"
G L2 Log Spherical harmonics 180.60 config="r2dm-g-kitti360-300k"
H L2 Log Fourier features 153.73 config="r2dm-h-kitti360-300k" (default)

KITTI-Raw (64x512, scan unfolding)

Config Loss Depth Positional Encoding FRD (T=1024) ↓ torch.load kwargs
H L2 Log Fourier features 207.31 config="r2dm-h-kittiraw-300k"