OK, so I'm new to this and doing the best I can. I'm trying to detect brightly colored Adirondack chairs around campus from drone imagery (0.03 m resolution). There are about 200 chairs in total. I digitized 50 training polygons, 25 for yellow chairs, 25 for blue. My workflow is 1) Export Training Data for Deep Learning, 2) Train Deep Learning Model, 3) Detect Objects using Deep Learning. I've been experimenting with different parameters such as tile size, stride size, and cell resolution. I'm generally satisfied with the results, many more errors of commission compared to omission. My question is this. WHY when I decrease the tile size in the Export Training Data tool (step 1) from say from 256 to 128 or 64, it will NOT permit me from change the tile size to match in the Detect Objects (step 3) without generating the following error: "Tile size should be greater than zero". No matter what I set the tile size to in step 1, I have to set it to 256 in step 3. And, therefore its generating the same exact same results. I'm not able to produce better or worse results?
