Vertex AI image classification models lose accuracy when being placed in a python dictionary

[Hillyu]

I have made a  model using vertex AI's image classification. Exported as EdgeTPU tflite model to my Raspberry pi 4 with Coral USB accelerator. When I used the Pycoral's example code https://github.com/google-coral/pycoral/blob/master/examples/classify_image.py  to run my model, I get a perfect prediction result. But when I passed them to a python dictionary in my script, the prediction accuracy is way off. https://github.com/hillyuyichu/Pycoral-python-API/blob/main/pycoral_classification.py 

Here is a screenshot of the prediction results on my python classification.py:

The label in row 1 is always the most active. The one in the last rows are the least active and most inaccurate.

ex: In picture 2, the label empty_pan barely ever cross 0.10 mark when it should have been more than 0.50

[comaro]

It seems Vertex AI is not supposed to be placed in a Python directory.

If this is impacting your application or your business, you can file a feature request using the following link. File the feature request, and they could assist you with the feature you are trying to implement.

[Hillyu]

Hi Comaro!

Thanks for the reply! I was able to find a way to work around it.

[comaro]

Could you share the workaround, Hillyu?