Closed faniry6 closed 4 years ago
For the early stage of the training, marching cube could fail due to noisy prediction of occupancy fields. To prevent the training process from being suspended by this, we intentionally use try-except to move on to next iterations. That's said, if you keep seeing this error during training, your network is likely not trained well to produce smooth and accurate occupancy fields.
Thanks for the explanation. For the moment we still get the same error at Epoch 23. I hope it will get better.
Thanks for the explanation. For the moment we still get the same error at Epoch 23. I hope it will get better. hi, how are you still getting this error in the current training situation, is there a good solution to share, looking forward to your reply
Nope, we never succeed to get the marching cube right :( Nothing new for the moment
@zyz-1998 Hi! Have you got some ideas about this issue?
Seem to be the same problem I got on PIFuHD https://github.com/facebookresearch/pifuhd/issues/85
Downgrading scikit-image fixed it pip install scikit-image==0.18.3
(same version of my Google Colab).
Downgrading scikit-image fixed it pip install scikit-image==0.18.3 Worked for me!
pip install scikit-image==0.19.2
also works, and perhaps explains the issue. It says: FutureWarning: marching_cubes_lewiner is deprecated in favor of marching_cubes. marching_cubes_lewiner will be removed in version 0.19
I'm thinking that pifuhd looks for marching_cubes_lewiner
Also, to fix issues with freeglut, I installed it from Conda using conda install -c anaconda freeglut
, while at the same time the PyOpenGL was the one from https://www.lfd.uci.edu/~gohlke/pythonlibs/#pyopengl
tested on python 3.10.4
Using your source code on geometry we generated, we got this error during the training "Error cannot marching cubes". This error appears when the iteration is finished, it restarts the iteration again. Is this expected or something is wrong with our settings?
Thanks again for sharing your source.