Reconstructed.tar.xz error

[Ibrahimatef]

Hi, Thanks for your great work I am trying to run your repo HumanViewSynthesis using infer_virtual.py so, I downloaded one video from panoptic data using this command ./scripts/getData.sh 171026_pose3 0 1 and followed your processing instructions and edited metadata.py according to the downloaded video but I get this error when trying to run bash list_reconstruction_3d.sh

{'calibration': PosixPath('/content/Panoptic/171026_pose3/cameras.json'),
 'max_iterations': 500,
 'min_views': 1,
 'pattern': '([0-9]{12})+',
 'poses_2d_dir': PosixPath('/content/Panoptic/171026_pose3/Poses/2D/BODY_135'),
 'poses_2d_suffix': '.tar.xz',
 'poses_3d': PosixPath('/content/Panoptic/171026_pose3/Poses/3D/BODY_135/Reconstructed.tar.xz'),
 'threshold': 0.25}
100% 7995/7995 [00:03<00:00, 2449.52it/s]
  0% 15/1079190 [00:00<11:49, 1520.34it/s]
Traceback (most recent call last):
  File "BODY_135.py", line 99, in <module>
    verbose=True,
  File "/content/Reconstruction3D/geometry/reconstruction.py", line 108, in perspective_reconstruction
    m_[:] = 0.0
IndexError: slice() cannot be applied to a 0-dim tensor.

Could you help me to fix it?

[GuillaumeRochette]

Hi @Ibrahimatef, In order to perform 3D reconstruction, you need to download videos from multiple viewpoints. The command ./scripts/getData.sh 171026_pose3 0 1 downloads the video of a single viewpoint. You need at least 2 views to run the 3D reconstruction algorithm, however the more the better usually, as estimated 2D keypoints are not pixel accurate. I recommend you to download all the 31 videos to reproduce the data used in the paper, using the following command ./scripts/getData.sh 171026_pose3 0 31, as indicated in the README. Best, Guillaume