metadata

license: apache-2.0

Preparing ISO

Datasets

We provide the OccScanNet dataset files here, but you should agree the term of use of ScanNet, CompleteScanNet dataset.

For a simplified way to prepare the dataset, you just download the preprocessed_data to ISO/data/occscannet as gathered_data and download the posed_images to ISO/data/scannet.

However, the complete dataset generating process is provided as followed:

OccScanNet

Clone the official MMDetection3D repository.

git clone https://github.com/open-mmlab/mmdetection3d.git ISO_mm

Swith to v1.3.0 version.

cd ISO_mm
git checkout v1.3.0

Download the ScanNet dataset following instructions and place scans directory as ISO_mm/data/scannet/scans.

:bulb: Note

Recommend you create a posed_images directory at data disk and link the scans directory and posed_images directory to data/scannet, then run the following command.

In this directory, extract RGB image with poses by running

python extract_posed_images.py --max-images-per-scene 100

:bulb: Note

Add --max-images-per-scene -1 to disable limiting number of images per scene. ScanNet scenes contain up to 5000+ frames per each. After extraction, all the .jpg images require 2 Tb disk space. The recommended 300 images per scene require less then 100 Gb. For example multi-view 3d detector ImVoxelNet samples 50 and 100 images per training and test scene.

Then obtained the following directory structure.

scannet
├── meta_data
├── posed_images
│   ├── scenexxxx_xx
│   │   ├── xxxxxx.txt
│   │   ├── xxxxxx.jpg
│   │   ├── intrinsic.txt
├── scans
├── batch_load_scannet_data.py
├── extract_posed_images.py
├── load_scannet_data.py
├── README.md
├── scannet_utils.py

Download original CompleteScanNet

The ground truth labels we used are from SCFusion. Ground truth is available at here.

The ground truth label should be placed as ISO_mm/data/completescannet/gt.

Reformulate CompleteScanNet

python preprocess_gt.py

The resulted directory is ISO_mm/data/completescannet/preprocessed.

Now, we obtained the following directory structure.

completescannet
├── gt
│   ├── scenexxxx_xx.ply
├── preprocessed
│   ├── scenexxxx_xx.npy
├── preprocess_gt.py
├── visualization.py

Create the OccScanNet

First, you should create a directories with name preprocessed_voxels and gathered_data in data disk and link them to the ISO_mm/data/occscannet.

python generate_gt.py

Now, we obtained the following directory structure.

occscannet
├── preprocessed_voxels
├── gathered_data
├── generate_gt.py
├── not_aligns.txt
├── wrong_scenes.txt
├── bad_scenes.txt
├── used_scannames.txt

OccScanNet-mini

The scenes we used in OccScanNet-mini is reflected in the config file.