tensorflow: Max and min for dynamic tensors should be recorded during calibration: Failed for tensor Shape

System information

• OS Platform and Distribution: Linux Ubuntu 18.04
• TensorFlow installed using pip, it’s actually tf-nightly-gpu
• TensorFlow version: tf-nightly-gpu==2.3.0.dev20200527

Command used to run the converter or code if you’re using the Python API

import numpy
import tensorflow as tf


def data_reader(size=900):
    return numpy.ones((size, 32, 86, 1), numpy.uint8) * 255


data = data_reader()
data = tf.cast(data, tf.float32)
repr_ds = tf.data.Dataset.from_tensor_slices((data)).batch(1)


def representative_data_gen():
    for input_value in repr_ds.take(100):
        yield [input_value]


converter = tf.lite.TFLiteConverter.from_saved_model("./exported")
converter.experimental_new_converter = True

converter.optimizations = [tf.lite.Optimize.DEFAULT]

converter.representative_dataset = representative_data_gen
converter.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]
converter.inference_input_type = tf.uint8
converter.inference_output_type = tf.uint8

tflite_model = converter.convert()

open("exported.tflite", "wb").write(tflite_model)

**The output from the converter invocation**

Traceback (most recent call last): File “…”, line 29, in <module> tflite_model = converter.convert() File “venv/lib/python3.6/site-packages/tensorflow/lite/python/lite.py”, line 920, in convert return super(TFLiteConverterV2, self).convert() File “venv/lib/python3.6/site-packages/tensorflow/lite/python/lite.py”, line 752, in convert self).convert(graph_def, input_tensors, output_tensors) File “venv/lib/python3.6/site-packages/tensorflow/lite/python/lite.py”, line 502, in convert constants.FLOAT, False) File venv/lib/python3.6/site-packages/tensorflow/lite/python/lite.py", line 349, in _calibrate_quantize_model inference_output_type, allow_float) File “venv/lib/python3.6/site-packages/tensorflow/lite/python/optimize/calibrator.py”, line 93, in calibrate_and_quantize np.dtype(output_type.as_numpy_dtype()).num, allow_float) RuntimeError: Max and min for dynamic tensors should be recorded during calibration: Failed for tensor Shape Empty min/max for tensor Shape

**Also, please include a link to the saved model or GraphDef**

https://drive.google.com/file/d/1dSDVWjM2fJbT71r-EiUqkwygtYr0a5aE/view?usp=sharing

**Failure details**
I am trying to convert this model by performing full integer quantization described [here](https://www.tensorflow.org/lite/performance/post_training_quantization#full_integer_quantization_of_weights_and_activations), but it fails as described above. If I attempt to convert it without a representative dataset it works ok, but the input/output tensors remain float32 and I need int8. Maybe I am missing something, but I'd like to know if this is solvable. Thanks in advance.