Wanting to try out some of the new deep learning tools in Pro 2.8. I have a clean install of 2.8, cloned Python environment and the deep learning modules installed by "conda install deep-learning-essentials".
I can successfully run the deep learning tools with CPU specified in the environment variables. But when I choose GPU I get an error box popup: "ArcGIS Pro has experienced a fatal authorization error and must close". Screenshot attached showing error and messages when running the Detect Objects tool.
I have ArcGIS Pro 2.8 Advanced and the Image Analyst extension licensed as a named user through ArcGIS Online. The same thing is occurring on two different machines, desktop with 5Gb Quadro P2000 and laptop with 8Gb GeForce RTX 2080. Both machines on Windows 10.
Am I just too quick off the mark trying to do deep learning in Pro 2.8?
Solved! Go to Solution.
Thanks for the update. I will look for the release and apply it as soon as it becomes available. In a meantime, I have reverted back to 2.7.4.
Good Day Shingo.
I also had to go back to 2.7 Pro version. However when i tried to install 2.7 libraries it says "A new version has been installed..". Could you tell me how to uninstall 2.8 DL Libraries. Thanks in advance.
To uninstall 2.8 DL Libraries, you can uninstall from standard windows add or remove program from the settings if you have installed from msi. If you are installing manually from conda deeplearning essential, you must have installed to your clone of arcgispro-py3 default python library, so you will just need to remove the environment and recreate a new environment once you have reinstalled 2.7.
With the recent 2.8.1, my "classify object using deep learning" issue has been resolved. It looks like 2.8.2 is about to be released, so I would stick with 2.8.1 and update the new version when it becomes available. If you need to install 2.7, you would need to uninstall Pro and reinstall 2.7 and get the installer from here.
The tool completed for small datasets (200 buildings); But for large datasets (1000 buildings) the tool was not completing even after 15 hours. In 2.7 the same dataset took 20 mins - whereas in 2.8 it is taking over 22 hours.