Inference for int8 efficientdet-d{$n} is not running unlike efficientdet-lite{$n}

[drahmad89]

I have followed the instruction provided by @fsx950223 to create a int8 quantized tflite model. The quantization was for weights and layers output. The tflite obtained from a efficientdet-d2 checkpoint was not able to run an inference (the script was running for ever without any output and it seems stuck somewhere). I tried to make inference with tflite converted from a efficientdet-lite0 and the inference went just fine.

Note: for the efficientdet-d2-int8.tflite the inference (using keras.eval_tflite) did not print out any error message and kept running forever.

[fsx950223]

tensorflow/tensorflow#40183

[drahmad89]

I let it run for 4 hours without any output

[drahmad89]

tensorflow/tensorflow#40183

The speed is not a problem. In case of d0 the inference is running but slow. but in case of d2 it is not running through one sample inference even if I wait hours.