Skip to content

Latest commit

 

History

History
68 lines (54 loc) · 2.2 KB

README.md

File metadata and controls

68 lines (54 loc) · 2.2 KB
Raw

Processed SceneNN Data

The processed SceneNN data can be downloaded from HERE.Run cat SceneNN.tar.* > SceneNN.tar to merge the files. Then skip to Step 4.

Prepare SceneNN-MV data for semantic segmentation test

Step 1. Download SceneNN oni data from HERE. Then download .ply data, .xml data and trajectory.log data from HERE. The data of each scene is structured as follows:

SceneNN
├── 005
│   ├── 005.ply                 /* the reconstructed triangle mesh  */
│   ├── 005.xml                 /* the annotation                   */
├── trajectory
│   ├── 005_trajectory.log      /* camera pose (local to world)     */
└── 005.oni                     /* the raw RGBD video               */

For simplicity and representativeness, we select 12 highly comprehensive and object-rich scenes , namely ['015', '005', '030', '054', '322', '263', '243', '080', '089', '093', '096', '011'].

Step 2. Process oni data by using the tool in the playback folder HERE.

Step 3. Process online data by:

python data_process.py

Step 4. Generate .pkl files by:

python tools/create_data.py scenenn --root-path ./data/scenenn-mv --out-dir ./data/scenenn-mv --extra-tag scenenn_mv

Final folder structure:

scenenn-mv
├── data_process.py
├── quality_check.py
├── scannet_utils.py
├── README.md
├── scannetv2-labels.combined.tsv
├── 005.oni
├── trajectory
│   ├── 005_trajectory.log
│   ├── 011_trajectory.log
├── 005
│   ├── depth
│   │   ├── depthxxxxx.png
│   ├── image
│   │   ├── imagexxxxx.png
│   ├── label
│   │   ├── xxxxx.npy
│   ├── point
│   │   ├── xxxxx.npy
│   ├── pose
│   |   ├── xxxxx.npy
|   ├── 005.ply 
|   ├── 005.xml 
|   ├── timestamp.txt
├── SceneNN_validate.pkl
└── scenenn_mv_infos_val.pkl