RoMe can also be applied for driving drowsiness reduction based on road surface element matching and alarming (Integration of Cabin and Driving). More details: https://fatigueview.github.io/
In configs/local_nusc.yaml
scene-0063, scene-0064, scene-0200, scene-0283
and scene-0109, scene-0508, scene-0523, scene-0821
respectively.In configs/local_kitti.yaml
torch==1.10.2+cu111
torchvision==0.11.3+cu111
torchaudio==0.10.2+cu111
pytorch3d==0.6.1
pymeshlab==2021.10
scipy opencv-py thon tqdm wandb python3.8
For wandb usage, please visit here.
Modify configs/local_nusc.yaml
base_dir
and image_dir
according to your folderclip_list
to train one scene or multiple scenes.run_local.sh
and then run sh run_local.sh
to start training.Modify configs/local_kitti.yaml
base_dir
and image_dir
according to your foldersequence
to choose which sequence to train.choose_point
and bev_x/y_length
to choose which sub area to train.run_local.sh
and then run sh run_local.sh
to start training.configs/nusc_eval.yaml
model_path
and pose_path
where your trained models saved.batch_size: 1
@misc{mei2023rome,
title={RoMe: Towards Large Scale Road Surface Reconstruction via Mesh Representation},
author={Ruohong Mei and Wei Sui and Jiaxin Zhang and Xue Qin and Gang Wang and Tao Peng and Cong Yang},
year={2023},
eprint={2306.11368},
archivePrefix={arXiv},
primaryClass={cs.CV}
}