I have my own pytorch, onnx, tf1.x(frozen_graph.pb), and tf2.x(saved_model.pb) models.
Each of those model outputs are exactly same what I wanted. (original model is