S3DIS dataset

1) Download the S3DIS dataset

2) Put the Stanford3dDataset_v1.2.zip to SoftGroup/dataset/s3dis/ folder and unzip

3) Preprocess data

cd SoftGroup/dataset/s3dis
bash prepare_data.sh

After running the script the folder structure should look like below

SoftGroup
├── dataset
│   ├── s3dis
│   │   ├── Stanford3dDataset_v1.2
│   │   ├── preprocess
│   │   ├── preprocess_sample
│   │   ├── val_gt

ScanNet v2 dataset

1) Download the ScanNet v2 dataset.

2) Put the downloaded scans and scans_test folder as follows.

SoftGroup
├── dataset
│   ├── scannetv2
│   │   ├── scans
│   │   ├── scans_test

3) Split and preprocess data

cd SoftGroup/dataset/scannetv2
bash prepare_data.sh

The script data into train/val/test folder and preprocess the data. After running the script the scannet dataset structure should look like below.

SoftGroup
├── dataset
│   ├── scannetv2
│   │   ├── scans
│   │   ├── scans_test
│   │   ├── train
│   │   ├── val
│   │   ├── test
│   │   ├── val_gt

STPLS3D dataset

1) Download the STPLS3D dataset

2) Put Synthetic_v3_InstanceSegmentation.zip to dataset/stpls3d and unzip

3) Preprocess data

cd SoftGroup/dataset/stpls3d
bash prepare_data.sh

SemanticKITTI dataset

1) Download the SemanticKITTI dataset

2) Unzip the downloaded data and put the sequences to dataset/kitti

3) The data structure should be as follows:

SoftGroup
├── dataset
│   ├── kitti
│   │   ├── sequences
│   │   |   |── 00
│   │   |   |   ├── calib.txt
│   │   |   |   ├── labels
│   │   |   |   ├── poses.txt
│   │   |   |   ├── times.txt
│   │   |   |   ├── velodyne
│   │   |   |── ...
│   │   |   |── 21