Tensorflow Lite interpreter `invoke()` hangs upon attempting int8 quantization with convertor

1. System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 16.04
• TensorFlow installation (pip package or built from source): pip package
• TensorFlow library (version, if pip package or github SHA, if built from source): 2.5.1

2. Code

Collab based on the provided reference collab:

https://gist.github.com/lucapericlp/8773916d530c2e9ca4d2dcd2b6a023c0

3. Failure after conversion

If the conversion is successful, but the generated model is wrong, then state what is wrong:

Model hangs indefinitely after invoking a converted model using uint8 quantization.

4. Details

I've performed some experiments & discovered the following which might prove useful:

• EfficientNetB3 & B4 convert & invoke successfully using only float16 quantization
• EfficentNetB3 & B4 convert successfully but fail invocation when using int8 quantization
• EfficientNetB0 converts & invokes successfully using float16 or uint8 quantization

Start with the linked reference Colab and reproduce the EfficientNetB3/B4 conversion and invocation behavior under float16, int8, and uint8 quantization. Compare the successful and hanging cases; done means the converted int8 or uint8 model invokes without hanging.