Upload preparations

All submissions must be uploaded using the python uploader. It must be placed under the upload folder and launched from there.

Upload folder

The upload folder must contain all the reconstruction files of at least one complete set of scenes (Intermediate, Advanced, or both). We only accept submissions of full sets. The set (Intermediate, Advanced or both) can be specified as a parameter of the python uploader. The evaluation can only be triggered once the uploader has successfully uploaded all files in the specified set(s) and validated the integrity of uploaded files.

The reconstruction of each scene should contain two files: a *.ply file of the reconstructed geometry and a *.log file of the recovered camera poses. Both files should be named after the according scene (e.g., Francis.ply, Francis.log; case sensitive). These files should be put directly under the upload folder, without sub-directories.

File format

Reconstructed geometry is stored in *.ply format. ASCII or binary. We treat reconstructed geometry as point clouds. If your reconstruction method produces a mesh, consider sampling it into a point cloud. See the τ column from table 1 of our paper for recommended point sampling density.

Recovered camera poses are stored in *.log. An example of a log file can also be seen here. We do not explicitly evaluate the camera poses. They are used only for the initialization of the evaluation process. Nevertheless, you should provide the complete set of camera poses. The number of items in the .log file should either match the number of images in the image set, or the number of frames in the video. If your method subsamples the input image sets or the video frames, you should interpolate to estimate the camera poses of other images/frames as well. The items in the .log file must be sorted according to image/frame sequence. In each data item, you can fill in any meta data (the three integer numbers of the first line of each item) and a 4-by-4 transformation matrix that transforms a point from camera space to global space.

Checklist

  • The upload folder contains the python uploader, the credentials file, and all reconstruction files from one or two complete sets
  • Each reconstruction contains a .ply file and a .log file, named after the scene, case sensitive
  • Reconstructed geometry is a (dense) point cloud
  • Number of recovered camera poses matches either the number of images in the image set, or the number of frames in the video.

Python uploader

Python uploader


Usage:

> python upload_t2_results.py [-h] [--group GROUP]

Example 1: upload intermediate and advanced reconstruction results
> python upload_t2_results.py --group both

Example 2: upload only intermediate results
> python upload_t2_results.py --group intermediate