I'd got a dog image, and I attempt to use a Pytorch model to classify, then convert it to ONNX and to Tensorflow type.
I expect to get a same classification result, since it's originally the same model. However, the Pytorch model outputs "tennis_ball" class while Tensorflow model outputs "Samoyed" class.
Based on the true label, I suppose the Tensorflow model is more accurate, but that result confused me. Why would the same model perform differently?
How could I do to make it static for the model performance?
I'd got a dog image, and I attempt to use a Pytorch model to classify, then convert it to ONNX and to Tensorflow type.
I expect to get a same classification result, since it's originally the same model. However, the Pytorch model outputs "tennis_ball" class while Tensorflow model outputs "Samoyed" class.
Based on the true label, I suppose the Tensorflow model is more accurate, but that result confused me. Why would the same model perform differently?
How could I do to make it static for the model performance?
Here is my working process: https://colab.research.google.com/drive/1pBU_pYdmBHPEPf2C5fRTGkP_cs5kxL0e?usp=sharing