Hello,
I've been experimenting a lot with the Classify Objects Using Deep Learning tool. I have .14 meter resolution imagery of coastal communities impacted by hurricanes, and a feature class of building footprints. I've been labeling the building features as either Undamaged (0), Damaged (1), or Missing (2), and training a model based on the imagery using the Train a Deep Learning Model script with the metadata type being Labeled Tiles. I've gotten the internal model metrics to be pretty good, as high as 90% accurate which I'm definitely happy about. Everything at this stage looks great. But when I run the model on new data within the same community, it always predicts every feature to be the same category with the same confidence level when clearly this is not the case. I'm completely out of ideas. What could be causing this?
The model is trained on 800-900 samples and I have ArcPro verison 2.9
Any help greatly appreciated
