TheCrazyT
commented
4 years ago Version 2 without downgrade of cuda/tensorflow(2.3.0 - using compatibility mode of v1) is here: https://colab.research.google.com/drive/1wmAJAxrD3ob2bR_WRUpS4B_-Lg-02b42?usp=sharing
It generates an output file (fig.png), but somehow the execution also crashes with an "Segmentation fault". Not shure what this is about, it is hard to debug.
Edit: Oh well figured out that it seems to crash inside the vtk-module since no X-server is running. I can live with that since vtk inside notebook/browser is not possible atm anyway.
Edit: Since vtk does not work for jupyter notebooks I created some test with "wiggle" animation:
https://colab.research.google.com/drive/1HDtdYcBTbUQbSJxakehPFTpmMcSyfxfG?usp=sharing (Was testing if AI could somehow be more clever than the human eye, that can't estimate the depth on that illusion. But looking at the depthmap ... the AI fails, too.)
Because I myself find this project impressive, I made a sample google colab notebook. I tried my best to make it work (and it kinda works).
I'm still trying to figure out to make it work without downgrading tensorflow. But its probably not that easy.
Here is the link to the notebook: https://colab.research.google.com/drive/1RQYj8r_zlTDcKxcBDOC_4CZGhMFxAYzR?usp=sharing Currently it does some dirty changes (like downgrading cuda and tensorflow and changing gcc header files), it is not advised to use it outside of google colab. Wish there would be a cleaner way, but I was not able to find one.
"vtk" is currently useless on google colab, but maybe there exist addons for displaying 3d models in webgl ...