Not able to generate IR for SSD-VGG-16-300 for voc dataset using MXNET

[Lakshmi_U_Intel]

Hi,

We are not able to generate IR for the OpenVino supported topology SSD-VGG-16-300 for voc dataset using MXNET.
Please find the below url for the pretrained  SSD-VGG-16-300 model.
(https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_convert_model_Convert_Model_From_MxNet.html)

OpenVino Version: openvino_2019.1.133(Linux)

mxnet-mkl version: 1.5.0

CPU Model: Intel(R) Xeon(R) Gold 6128 CPU

The command used for running the model optimizer is:
python mo_mxnet.py --input_model ~/vgg16_ssd_300_voc0712_trainval/ssd_300-0000.params --input_shape [1,3,300,300] --log_level=DEBUG
 
The error during model optimization is  given below:
 
[ WARNING ]  Node attributes: {'symbol_dict': {'op': '_contrib_MultiBoxTarget', 'name': 'multibox_target', 'attr': {'ignore_label': '-1', 'minimum_negative_samples': '0', 'negative_mining_ratio': '3', 'negative_mining_thresh': '0.5', 'overlap_threshold': '0.5', 'variances': '(0.1, 0.1, 0.2, 0.2)'}, 'inputs': [[147, 0, 0], [148, 0, 0], [133, 0, 0]]}, '_in_ports': set(), '_out_ports': set(), 'kind': 'op', 'name': 'multibox_target', 'type': '_contrib_MultiBoxTarget', 'op': '_contrib_MultiBoxTarget', 'infer': None, 'precision': 'FP32'}

[ ERROR ]  Unexpected exception happened during extracting attributes for node label.

Original exception message: Operation '_contrib_MultiBoxTarget' not supported. Please register it as custom op.

For more information please refer to Model Optimizer FAQ (<INSTALL_DIR>/deployment_tools/documentation/docs/MO_FAQ.html), question #86.

[Shubha_R_Intel]

Dear  U Menon, Lakshmi,

I have answered your previous IDZ post which asks a similar question.

Thanks,

Shubha

[Lakshmi_U_Intel]

Hi Shubha,

Thanks for your reply.

My post (https://software.intel.com/en-us/forums/computer-vision/topic/814952) was regarding the GluonCV MXNET model.
But when I tried with the OpenVino supported topology SSD-VGG-16-300 using MXNET (Not using GluonCV) it was also giving the same custom operation related error.(Please find the above post for the error trace).

Please find the below url for the MXNET SSD-VGG-16-300 model.

(https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_convert_model_Convert_Model_From_MxNet.html)

I wanted to do inferencing on OpenVino with the pretrained SSD-VGG-16-300 model using MXNET.

Is there any option to use any of the object detection topologies using MXNET in OpenVino R1?

Thanks,

Lakshmi.

[Shubha_R_Intel]

Dear U Menon, Lakshmi ,

When I look at these two places I am not finding the MXNET SSD-VGG-16 model:

https://github.com/zhreshold/mxnet-ssd/tree/master/model

https://mxnet.apache.org/model_zoo/index.html

The links on our documentation seem to point to a bad site (I will file a bug on this). 

Please let me know where you got the *.json and *.params for the model. 

Thanks,

Shubha

[Lakshmi_U_Intel]

Hi Shubha,

I had downloaded the .json and .params file from

 https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_convert_model_Convert_Model_From_MxNet.html 

It will download a zip file vgg16_ssd_300_voc0712_trainval.zip which contains ssd_300-0000.params and ssd_300-symbol.json

As you mentioned, in https://github.com/zhreshold/mxnet-ssd/tree/master/model and https://mxnet.apache.org/model_zoo/index.html there is no model for SSD-VGG-16 model.

Thanks,

Lakshmi.

[Shubha_R_Intel]

Dear U Menon, Lakshmi ,

OK that is reasonable. You got ssd_300-0000.params and ssd_300-symbol.json from the Model Optimizer recommended place. I see that you are using an extremely ancient version of OpenVino  : openvino_2019.1.133(Linux). Can you kindly upgrade to 2019R2 which we just released this week and try again ? Please report your status here.

Thanks,

Shubha

[Lakshmi_U_Intel]

Hi Shubha,

I tried with 2019R2 on both Windows and Linux Machine and I am getting the same error which I got while trying with 2019R1. I tried 2019R2 on both windows and Linux machine and getting the same error.

python mo_mxnet.py --input_model ~/vgg16_ssd_300_voc0712_trainval/ssd_300-0000.params --input_shape [1,3,300,300] --log_level=DEBUG

Please find the error trace.

[ WARNING ]  Node attributes: {'symbol_dict': {'op': '_contrib_MultiBoxTarget', 'name': 'multibox_target', 'attr': {'ignore_label': '-1', 'minimum_negative_samples': '0', 'negative_mining_ratio': '3', 'negative_mining_thresh': '0.5', 'overlap_threshold': '0.5', 'variances': '(0.1, 0.1, 0.2, 0.2)'}, 'inputs': [[147, 0, 0], [148, 0, 0], [133, 0, 0]]}, '_in_ports': {}, '_out_ports': {}, 'kind': 'op', 'name': 'multibox_target', 'type': '_contrib_MultiBoxTarget', 'op': '_contrib_MultiBoxTarget', 'infer': None, 'precision': 'FP32'}
[ ERROR ]  Unexpected exception happened during extracting attributes for node label.
Original exception message: Operation '_contrib_MultiBoxTarget' not supported. Please register it as custom op.
 For more information please refer to Model Optimizer FAQ (https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_Model_Optimizer_FAQ.html), question #86.

Thanks,

Lakshmi

[Shubha_R_Intel]

Dear U Menon, Lakshmi, as I stated in your other post, I reproduced your bug regarding mo_mxnet.py handling of vgg16_ssd_300 and have filed a high priority bug on your behalf.

Sorry for the trouble,

Thanks,

Shubha

[Lakshmi_U_Intel]

Thanks Shubha. It would be great if you could let us know once the issue is resolved so that I can try out mo_mxnet.py with vgg16_ssd_300

Thanks,

Lakshmi.

[singh__yogender]

hi Shubha,

we are not able to generate IR for openvino supported model ssd_300_vgg16  which trained on custom dataset using voc as a data format.

openvino version : openvino_2019.3.376 

processor : Intel® Core™ i7-8750H CPU @ 2.20GHz × 12 

Command used for the optimization :

sudo python3 mo_mxnet.py --input_model /home/omen/Desktop/incubator-mxnet/example/ssd/model/ssd_vgg16_reduced_300-0000.params --input_symbol /home/omen/Desktop/incubator-mxnet/example/ssd/model/ssd_vgg16_reduced_300-symbol.json --legacy_mxnet_model

i get error during model optimization

[ ERROR ]  Operation '_contrib_MultiBoxTarget' not supported. Please register it as custom op.

Unexpected exception happened during extracting attributes for node label.
Original exception message: Operation '_contrib_MultiBoxTarget' not supported. Please register it as custom op. 

For more information please refer to Model Optimizer FAQ (https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_Model_Optimizer_FAQ.html), question #86.

it would be great if issue is resolved ASAP.

Thanks