I have been working with Deep Learning and ArcPro--fantastic. So far, so good. I've run through my work creating the training samples by hand, up to several hundred separate polygons for training. Now I would like to attempt to automate the creation of the training samples somewhat.
The Question: I have an input dataset of points that I would like to buffer and then use these polygon areas as training samples with a class value of 1. I can hand digitize these (as mentioned above and following the tutorials, thanks for these resources ESRI), but would now like to determine if I can buffer the points and them import all of these buffers directly with one class value (true) already in their attribute table for deep learning. I can import such as buffer polygon file into the training sample manager, but each single polygon gets a unique class value, which I don't want. I pre-defined the class variable and value in the polygon attribute table (creating a new attribute variable called Class, set it to 1 within the attribute table manager), but this is ignoring during the import operation within the training sample manager. I still end up with every single polygon having a unique class value.
I've checked the ArcPro documentation without success. Any suggestions would be greatly appreciated! Thank you.
Hello William, I am facing the same problem you described. I already have a huge data set with thousands of training samples, but the import does not really work properly. Would you share your experience on the correct setup of the attributes? Or does someone else have found a solution?
Many thanks in advance!
I found a solution through generating a small training sample data set. The attribute table shows the following fields: Classname (text), Classvalue (Long), Classcode (Text, e.g. Code_1). Additional columns: RED, GREEN, BLUE (all Long) define the color for the classes displayed in the map.