Unable to find the overall accuracy of TFv2_TF_TRT FP32, 16 and 08 based classification model

[user164502]

I am running Keras-Tensorflow version 2.0 based image classification model on Jetson Xavier.
I know that we can find the overall accuracy of a CNN based image classification model in Tensorflow by
results = model.evaluate(test_images, verbose=0)
print(" Loss: {:.5f}".format(results[0]))
print(“Accuracy: {:.3f}%”.format(results[1] * 100))

But how can we find the overall accuracy when we optimize an image classification model with TFv2_TF_TRT FP32, 16 and 08 on Xavier?

I want to know the overall accuracy of optimized image classification model on Jetson Xavier. But I could not find any code or function. Therefore, I need help.

[AastaLLL]

Hi,

model.evaluate(.) requires a Keras model but TF-TRT is a TensorFlow model.
The simplest way is to do the inference and compare the output with the testing label manually.

Some TF-TRT inference examples can be found in the below link:
https://docs.nvidia.com/deeplearning/frameworks/tf-trt-user-guide/index.html#worflow-with-savedmodel

Thanks.