I do need to increase the accuracy I am getting. Not quite where the model needs to be yet.
Does Prediction try to compare the whole target image with each of the training images or does it grab the largest objects in the image and attempt to match them with a similar Training image?
If Prediction attempts to match the whole target image with Training images, background would seem to be more important. In my case, each picture taken by the camera typically has a much more area occupied by background than the desired object occupies. So I was attempting to "help" by distilling the Training images down to just the object. I don't know of a way to eliminate the background from the target image because the target object can be anywhere on the image. Worse: the lighting conditions are constantly changing. Also unfortunate is that the background consists of driveway, lawn, trees, shadows, etc. And those items/noise are in various locations about the target object.
Thanks for your patience with this process Jared.