[Tensorflow] Operator FakeQuantWithMinMaxVars not implemented

[titikid]

I tried to run quantized models by tensorflow and i got this error. Could anyone give me some hints to implement this operator: FakeQuantWithMinMaxVars?

Traceback (most recent call last):
File “from_tensorflow.py”, line 55, in
sym, params = nnvm.frontend.from_tensorflow(graph_def)
File “/usr/local/lib/python2.7/dist-packages/nnvm-0.8.0-py2.7.egg/nnvm/frontend/tensorflow.py”, line 1202, in from_tensorflow
sym, params = g.from_tensorflow(graph)
File “/usr/local/lib/python2.7/dist-packages/nnvm-0.8.0-py2.7.egg/nnvm/frontend/tensorflow.py”, line 1005, in from_tensorflow
op = self._convert_operator(node.op, inputs, attr, graph)
File “/usr/local/lib/python2.7/dist-packages/nnvm-0.8.0-py2.7.egg/nnvm/frontend/tensorflow.py”, line 1169, in _convert_operator
raise NotImplementedError(“Operator {} not implemented.”.format(op_name))
NotImplementedError: Operator FakeQuantWithMinMaxVars not implemented.

[zhongjun]

Has this problem been solved?

[ehsanmok]

I think it’s part of the RFC. Am I right @FrozenGene ?

[FrozenGene]

Yeah. FakeQuantWithMinMaxVars op will be used for quantization min / max information of op. After quantization (i.e. after using tflite_converter or toco), this op will not be existed in tflite model. So, we shouldn’t import this model in Tensorflow frontend, should use TFLite frontend. I will support importing TFLite int8 model in TVM dev 0.6, then we can handle this condition.