ScanNet and ScanNet200 Data Preparation

Look at data_preparation/dataset/globals_dirs.py and change the folder paths where you would like to store the data.

Downloading ScanNet Mesh/PointCloud Data

Download ScanNet v2 data from HERE. Let DATA_ROOT be the path to folder that contains the downloaded annotations. Under DATA_ROOT there should be a folder scans. Under scans there should be folders with names like scene0001_01. We need _vh_clean_2.ply, _vh_clean_2.0.010000.segs.json, _vh_clean_2.labels.ply, _vh_clean.aggregation.json. We provide a helper download script: download_scannet_files.py. This file downloads only the relevant portion of the scannet dataset we need. You still need to download the download-scannet-v2.py after filling the ScanNet agreement form before using this helper script. Additionally, you might need to do some modifications to the download-scannet-v2.py, for eg. removing the input("") that requires a manual keyboard input for each scene.

For ScanNet, execute

python data_preparation/scannet/scannet_preprocessing.py preprocess --data_dir PATH_TO_RAW_SCANS --save_dir SAVE_DATA

Add --scannet200 True for ScanNet200

Make sure to change SCANNET_DATA_DIR in odin/config.py to the `SAVE_DATA/train_validation_database.yaml'

Similarly, change SCANNET200_DATA_DIR in odin/config.py to the `SAVE_DATA/train_validation_database.yaml'

We provide preprocessed RGB-D data (~80G) for all scenes. You can downloading it using gdown in the data directory.

gdown --id 1Xq84J9Gl9CVns_4Q0gDBxcPoA7hSf-WY

You can skip this if you just want to use our preprocessed RGB-D data

First download the .sens files as well by using --type .sens argument with the scannet download script.
Execute the following script (make sure to change the data directory paths in the script)

 python data_preparation/scannet/preprocess_sens.sh

For ScanNet, execute:

python data_preparation/scannet/scannet2coco.py

For ScanNet200, add --scannet200 to the above command