Skip to content

Latest commit

 

History

History

datagen

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
SUNCGScan
SUNCGScan
 
 
ScanReal
ScanReal
 
 
SceneSampler
SceneSampler
 
 
fileLists
fileLists
 
 
README.md
README.md
 
 

README.md

Data Generation

Data generation code is developed under VS2013, and we use mLib and mLibExternal to generate our data, for setting up libraries, please checkout respective websites. Datasets used in this project are from ScanNet and SUNCG.

ScanNet

  • Code for ScanNet data generation is located in ScanReal
  • Setup the paths in ScanReal/zParameters.txt
  • Generate intermediate data for ScanNet from ScanReal
  • Setup the paths in SceneSampler/zParameters.txt
  • Comment out the line for ScanNet data.
  • Generate final data including chunks used in training and scenes used in testing from SceneSampler
  • Generate 2D images used for projection layer: enter folder ScanReal/python and run
 python prepare_2d_data.py --scannet_path /ScanNet/public/v1/scans --output_path ./scannetv2_images --label_map_file /ScanNet/public/v2/tasks/scannetv2-labels.combined.tsv

SUCNG

  • Code for ScanNet data generation is located in SUNCGScan
  • Setup the paths in SUNCGScan/zParameters.txt
  • Generate intermediate data and 2D images for SUNCGScan from SUNCGScan
  • Setup the paths in SceneSampler/zParameters.txt
  • Uncomment the line for SUNCG data.
  • Generate final data including chunks used in training and scenes used in testing from datagen/SceneSampler

Viusalizations

  • You can visualize intermediate data in C++ code by setting s_bDebugOut = true in zParameters.txt file.
  • You can also generate visualizations of the final data (chunks or scenes) in voxels by running python tools/visualization.py --path /final_data/ --mode data. Example visualization: 3dsis3dsis