Point clouds

  1. Downloading

    Building Name (Building Code) Floor # (Floor Code) Download Link Number of Scans Area (ft^2) % Correct
    Center (DUC) Floor 1 (DUC1) LRZ (15GB)
    ZIP (24GB)
    50 ~100,000 98%
    Center (DUC) Floor 2 (DUC2) LRZ (23GB)
    ZIP (39GB)
    80 ~100,000 100%
    Hall (CSE) Floor 3 (CSE3) LRZ (2.4GB)
    ZIP (3.9GB)
    7 ~8,000 85%
    Hall (CSE) Floor 4 (CSE4) LRZ (24GB)
    ZIP (40GB)
    75 ~19,000 93%
    Hall (CSE) Floor 5 (CSE5) LRZ (22GB)
    ZIP (36GB)
    65 ~20,000 66%
  2. Decompressing

    The point-clouds are available as LRZ or ZIP compressed PTX files. We provide lrzipped to accommodate those with slow Internet connections and/or data-caps. Here is a quick guide to getting lrzip and decompressing the files. However, we will not provide any additional support in terms of using/obtaining lrzip and encourage you to simply use the zipped versions if possible.

    1. Install lrzip

      • Debian/Ubuntu

        sudo apt-get install lrzip

      • Fedora

        sudo yum install lrzip

      • OS X

        brew install lrzip

      • Windows

        Cygwin provides a version of lrzip

    2. Decompress

      lrunzip *.lrz

  3. Usage

    Files are named as such: <Building Code>_scan_<Scan_ID>.ptx

    Information on the PTX format can be found here. We only utilize the rows and columns field and transformation matrices are in separate files however.

    The points in the PTX file constitute as column-major panorama with the number of rows and columns specified in the PTX header.

Transformation Matrices

  1. Download

    Floor Code Download Link
    DUC1 DUC1 Alignment
    DUC2 DUC2 Alignment
    CSE3 CSE3 Alignment
    CSE4 CSE4 Alignment
    CSE5 CSE5 Alignment
  2. Decompress

    unzip alignment.zip

  3. What you get

    Transformation matrices are provided in two places, in all_transformations.txt formated as such:

      Scan_ID
    
      Before general icp:
      Before GICP 4x4 Transformation Matrix
    
      After general icp:
      After GICP 4x4 Transformation Matrix
    

    The Before GICP 4x4 Transformation Matrix is the result of our algorithm. We have then used GICP to improve the fine alignment. Note that the "Before" and "After" GICP matrices may be the same in the case that GICP failed

    Scan_ID corresponds to the PTX filename.

    Transformation matrices are also in individual files in the transformations folder withe the naming scheme <Building Code>_trans_<Scan_ID>.txt

    The Scan_ID of known incorrectly placed scans are in known_incorrect.txt

  4. Usage

    Transformation matrices can be used as such. Let p be a point from a PTX file, p' be the coordinate of that point in the aligned coordinate system, and T be the 4x4 transformation matrix for that scan.

    Using homogeneous coordinates

    p' equal-up-to-scale T * p
    

    Using Eigen

    p' = (T * p.homogeneous()).eval().hnormalized()
    

Floorplan and Scale

The scale for DUC1 and DUC2 has included in their respective alignment.zip's as scale.txt. The floorplans can be found here.

License

This Wustl Indoor RGBD dataset is made available under the Open Database License: http://opendatacommons.org/licenses/odbl/1.0/. Any rights in individual contents of the database are licensed under the Database Contents License: http://opendatacommons.org/licenses/dbcl/1.0/

If you use our data, please cite our paper:

@inproceedings{wijmans17rgbd,
  author = {Erik Wijmans and
            Yasutaka Furukawa},
  title = {Exploiting 2D Floorplan for Building-scale Panorama RGBD Alignment},
  booktitle = {Computer Vision and Pattern Recognition, {CVPR}},
  year = {2017},
  url = {http://cvpr17.wijmans.xyz/CVPR2017-0111.pdf}
}