The json file is empty from tensorflow to IR about mmdnn

[kitstar]

Hi, could you share the model files and output logs?

[JackieLeeTHU11]

@kitstar
The commond is
python -m mmdnn.conversion._script.convertToIR -f tensorflow -d smallnet_v3 -n smallnet_v3_CSattention/smallnet_v3.ckpt.meta --dstNodeName 3dface_smallnet_v3/e_Regresion/e_regresion_layer/weights

the output log is:
Parse file [smallnet_v3_CSattention/smallnet_v3.ckpt.meta] with binary format successfully. Tensorflow model file [smallnet_v3_CSattention/smallnet_v3.ckpt.meta] loaded successfully. IR network structure is saved as [smallnet_v3.json]. IR network structure is saved as [smallnet_v3.pb]. Warning: weights are not loaded.

The model file is models, and the password is cxn7

[kitstar]

Hi, your dstNode is squeeze_logits, not 3dface_smallnet_v3/e_Regresion/e_regresion_layer/weights. Please try it again. And there are some layers conversion not completed. I will inform you if it finishs.
Thanks.

[JackieLeeTHU11]

@kitstar Hi, you are right. However, when I set dstNode with squeeze_logits, there is a bug. And the log is:

File "mmdnn/conversion/tensorflow/tensorflow_parser.py", line 281, in gen_IR
func(current_node)
File "mmdnn/conversion/tensorflow/tensorflow_parser.py", line 360, in rename_Conv2D
IR_node = self._convert_identity_operation(source_node, 1, 'Conv')
File "mmdnn/conversion/tensorflow/tensorflow_parser.py", line 387, in _convert_identity_operation
TensorflowParser._copy_and_reop(source_node, IR_node, new_op)
File "mmdnn/conversion/tensorflow/tensorflow_parser.py", line 303, in _copy_and_reop
assign_IRnode_values(IR_node, kwargs)
File "mmdnn/conversion/common/utils.py", line 45, in assign_IRnode_values
assign_attr_value(IR_node.attr[name], val)
File "mmdnn/conversion/common/utils.py", line 40, in assign_attr_value
raise NotImplementedError('AttrValue cannot be of %s' % type(val))
NotImplementedError: AttrValue cannot be of <type 'unicode'>

[kitstar]

It is a compatibility problem between py2 and py3. Fixed in newest code.

[JackieLeeTHU11]

Ok, I will try the newest code. Thanks for your work!