README.md

Prepare datasets:

KITTI-360: https://www.cvlibs.net/datasets/kitti-360/download.php only KITTI-360--data_3d_raw is used

Seeing Through Fog: https://light.princeton.edu/datasets/automated_driving_dataset

Prepare environment:

conda env create -f environment.yml

Pretrain and finetune:

set train_model at 'train' or 'finetune'

accelerate launch --mixed_precision 'fp16' --dynamo_backend 'no' train.py

Generate:

python generate.py

The weight file can be downloaded at this link: https://pan.baidu.com/s/17Z7ZgmTDuJ5thlxIn97PiQ, Password: 7788

The generated and labeled data can be found in https://pan.baidu.com/s/1_waBH02ZXpSlEKFA-o5_bw, password: 7878

The label tool is: https://github.com/ch-sa/labelCloud