This is a folder to contains a toy dataset used for LiDAR-MOS
Please use the recommended data structure as follows:
data
├── sequences
│ └── 08
│ ├── calib.txt # calibration file provided by KITTI
│ ├── poses.txt # ground truth poses file provided by KITTI
│ ├── velodyne # velodyne 64 LiDAR scans provided by KITTI
│ │ ├── 000000.bin
│ │ ├── 000001.bin
│ │ └── ...
│ ├── clean_scans # clean scans after applying our MOS results as masks
│ │ ├── 000000.bin
│ │ ├── 000001.bin
│ │ └── ...
│ ├── labels # ground truth labels provided by SemantiKITTI
│ │ ├── 000000.label
│ │ ├── 000001.label
│ │ └── ...
│ └── residual_images_1 # the proposed residual images
│ ├── 000000.npy
│ ├── 000001.npy
│ └── ...
├── predictions_salsanext_residual_1_valid # MOS results using SalsaNext with 1 residual images
│ └── sequences
│ └── 08
│ └── predictions
│ ├── 000000.label
│ ├── 000001.label
│ └── ...
├── predictions_salsanext_sem_valid # SalsaNext semantic segmentation predictions
│ └── sequences
│ └── 08
│ └── predictions
│ ├── 000000.label
│ ├── 000001.label
│ └── ...
└── model_salsanext_residual_1 # MOS pretrained model using SalsaNext with 1 residual images
├── arch_cfg.yaml
├── data_cfg.yaml
└── SalsaNext_valid_best