Intel® Distribution of OpenVINO™ Toolkit
Community support and discussions about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all things computer vision-related on Intel® platforms.

MobileNet V1 Convert to OpenVINO IR Error

Gouveia__César
New Contributor I
231 Views

Hi,

I tried to convert MobilenetV1 on OpenVINO and I got this error:

[ ERROR ]  Cannot infer shapes or values for node "Preprocessor/map/TensorArrayStack/TensorArrayGatherV3".
[ ERROR ]
[ ERROR ]
[ ERROR ]  It can happen due to bug in custom shape infer function <function TensorArrayGather.array_infer at 0x000001A80D374510>.
[ ERROR ]  Or because the node inputs have incorrect values/shapes.
[ ERROR ]  Or because input shapes are incorrect (embedded to the model or passed via --input_shape).
[ ERROR ]  Run Model Optimizer with --log_level=DEBUG for more information.
[ ERROR ]
Stopped shape/value propagation at "Preprocessor/map/TensorArrayStack/TensorArrayGatherV3" node.

 

Does anyone have the same issue?

Thanks,

César.

0 Kudos
4 Replies
Luis_at_Intel
Moderator
231 Views

Hi Gouveia, César,

The error might be related to an unsupported/custom layer found in your model, which can't be converted by the Model Optimizer. Please share the full command used to convert the model, share more details about the model (custom trained?), your OpenVINO version, OS, etc.. 

If possible share the model files as well and we can help with the conversion.

Regards,

Luis

Gouveia__César
New Contributor I
231 Views

Hi Luis, thank you very much for your response!

Sorry for not giving all the informations in the previous post, I left the office in a hurry so I got almost no time to make the post :D

So all the information:

OpenVINO version: openvino_2019.3.379;

OS: Windows 10;

DL Framework: TensorFlow;

Full log debug file: Attached.

If possible share the model files as well and we can help with the conversion

I would really appreciate if you could help me convert this model, you can send me a PM and I can send you the model and all the files you need to recreat the issue.

Thanks,

César.

Gouveia__César
New Contributor I
231 Views

Hi Luis,

Please send me a PM so I can send you the model and all the files you need to recreat the issue.

Thanks,

César.

Gouveia__César
New Contributor I
231 Views

Hi Luis,

So by looking at the logs file:

[ 2020-02-03 16:51:50,145 ] [ DEBUG ] [ infer:129 ]  --------------------
[ 2020-02-03 16:51:50,145 ] [ DEBUG ] [ infer:130 ]  Partial infer for Preprocessor/map/TensorArrayStack/range
[ 2020-02-03 16:51:50,146 ] [ DEBUG ] [ infer:131 ]  Op: Range
[ 2020-02-03 16:51:50,146 ] [ DEBUG ] [ infer:142 ]  Inputs:
[ 2020-02-03 16:51:50,147 ] [ DEBUG ] [ infer:32 ]  input[0]: shape = [], value = 0
[ 2020-02-03 16:51:50,147 ] [ DEBUG ] [ infer:32 ]  input[1]: shape = [], value = 1
[ 2020-02-03 16:51:50,148 ] [ DEBUG ] [ infer:32 ]  input[2]: shape = [], value = 1
[ 2020-02-03 16:51:50,148 ] [ DEBUG ] [ infer:144 ]  Outputs:
[ 2020-02-03 16:51:50,150 ] [ DEBUG ] [ infer:32 ]  output[0]: shape = [1], value = [0]
[ 2020-02-03 16:51:50,152 ] [ DEBUG ] [ infer:129 ]  --------------------
[ 2020-02-03 16:51:50,159 ] [ DEBUG ] [ infer:130 ]  Partial infer for Preprocessor/map/TensorArrayStack/TensorArrayGatherV3
[ 2020-02-03 16:51:50,159 ] [ DEBUG ] [ infer:131 ]  Op: TensorArrayGatherV3
[ ERROR ]  Cannot infer shapes or values for node "Preprocessor/map/TensorArrayStack/TensorArrayGatherV3".
[ ERROR ]
[ ERROR ]
[ ERROR ]  It can happen due to bug in custom shape infer function <function TensorArrayGather.array_infer at 0x000001A25A018510>.
[ ERROR ]  Or because the node inputs have incorrect values/shapes.
[ ERROR ]  Or because input shapes are incorrect (embedded to the model or passed via --input_shape).
[ 2020-02-03 16:51:50,170 ] [ DEBUG ] [ infer:196 ]  Node "Preprocessor/map/TensorArrayStack/TensorArrayGatherV3" attributes: {'pb': name: "Preprocessor/map/TensorArrayStack/TensorArrayGatherV3"

The layer/operation who gives the error is called TensorArrayGatherV3, I looked for it in the supported OpenVINO operations for TensorFlow and I found out that it is only supported when it is fused to the TensorIterator layer. So I think that's the problem because mine is fused with a TensorArrayStack (I think). Can you give me some help solving this problem? Can you send me a PM so that I can share the model privately?

Thanks,

César.

Reply