Label bounding box for patches sharing the same predicted feature

Question

I ran a patch-based convolutional neural network and I want to ‘morph’ neighbouring patches with the same predicted class into larger patch. In the example table below fa