Format for reconstruction task: the results should be reported as follows. For each image in the testset, the reconstructed voxels are saved in a file whose base filename is the same as the model name, with an extension of '.mat'. The voxels are saved as three diemsnion tensors with variable name 'voxel' inside '.mat' file. For example:

Input file:
  034670/11.png

Reconstruction:
  034670_11.mat

Each team should create a zip archive containing the reconstructed voxels for all test set shape. The archive should be named as AuthorLastName.zip and the directory inside should be named as: AuthorLastName_voxels.

Format for segmentation task: point label prediction for test set models should be organized in the same way as the provided training and validation sets. Each team should create a zip archive containing the predicted point labels. The archive should be named as AuthorLastName.zip and the directory inside should be named as: AuthorLastName_labels.

For both tasks: the participants should send the MD5 checksum code and a download link (Dropbox, SkyDrive or other downloadable location) to their zip file to the organizers through email at shapenetchallenge.iccv17@gmail.com. Once the organizers have successfully downloaded the results and verified the MD5, a confirmation email will be sent to each team.