Skip to content

Latest commit

 

History

History
65 lines (54 loc) · 1.92 KB

data.md

File metadata and controls

65 lines (54 loc) · 1.92 KB
Raw

Processed data

The processed data can be downloaded from HERE.
Run cat mmdet_xxx.tar.* > mmdet_xxx.tar to merge the files.
Then set the variable data_root in configs to the path of directory that contains the .pkl files.

ScanNet-md40

Step 1. Download ScanNet v2 data HERE. Link or move the scans/ folder to DSPDet3D/data/ScanNet-md40/.

Step 2. Following mmdetection3d to process the data.
First extract point clouds and annotations by running:

cd DSPDet3D/data/ScanNet-md40
python batch_load_scannet_data.py

Then use tools/create_data.py from mmdetection3d to generate .pkl files.

python DSPDet3D/tools/create_data.py scannet --root-path DSPDet3D/data/ScanNet-md40 --out-dir DSPDet3D/data/ScanNet-md40 --extra-tag scannet

Final folder structure:

ScanNet-md40
├── instance_mask/
├── points/
├── seg_info/
├── semantic_mask/
├── scannet_infos_train.pkl
├── scannet_infos_val.pkl
└── ...

TO-SCENE-down

Step 1. Download TO-SCENE dataset (TO_ScanNet version) from HERE. Download meta_data from HERE and move it into TO-scannet/.

The folder structure:

TO-scannet
├── meta_data/
├── train/
├── val/
└── test/

Link or move this folder to DSPDet3D/data/TO-SCENE-down/.

Step 2. Process the data by running:

cd DSPDet3D/data/TO-SCENE-down
python to-scannet_converter.py

Final folder structure:

TO-SCENE-down
├── instance_mask/
├── points/
├── seg_info/
├── semantic_mask/
├── toscene_infos_train.pkl
├── toscene_infos_val.pkl
└── ...