In the Train Deep Learning Model there is the option to specify a pre-trained Model. I am thinks about creating multiple Training Data sets and then process them one at a time through the Train Deep Learning Model tool.
For example we Exported Training Data for Deep Learning from one of ESRI's provided examples. It ran for 254 hours and produced 200,000 samples. Could we divide the training sample data into, lets say 10 areas each to be process the by the Export Training Data for Deep Learning tool, using identical parameters of course. (divide and conqueror using multiple computers.)
Using this approch we could train the first 20,000 training samples and when that is done run the next 20,000 training samples using the option of specifying a pre-trained Model. Is this practical? Has anyone tried this? Any downsides to this approach?
Thanks,
