"Train Point Cloud Classification Model" failed on cpu-only computer #113
Description
I installed ArcGIS Pro on a computer that does not have NVIDIA GPU and was able to install the "deep-learning-essentials" package into a clone of the default arcgispro-py3 environment.
I then ran Training Point Cloud Classification Model geoprocessing tool with RandLA-Net as the Model Architecture. I set ProcessorType to CUP in Environments. Still, the tool run failed, with an error saying "ERROR 050160: The specified GPU does not exist or is not supported."
I reviewed the python scripts tool.script.execute.py (in "C:\Users\xxxx\AppData\Local\Programs\ArcGIS\Pro\Resources\ArcToolBox\toolboxes\3D Analyst Tools.tbx\TrainPointCloudClassificationModel.tool") and _RandLANet.py and _rand_lanet_utils.py (both in "C:\Users\xxxx\AppData\Local\ESRI\conda\envs\arcgispro-py3-cpu\Lib\site-packages\arcgis\learn\models"), but did not found any place that's hard coded to use CUDA.
Why the tool is till trying to use GPU? From tool.script.execute.py I see "Point Transformer V3" model architecture explicitly requires CUDA, but there is no explicit GPU requirement for RandLA-Net.
@cleebp @BzGEO Any suggestions regarding how to make the tool work on my cpu-only computer would be greatly appreciated !!! Thank you!