Run deeplabv3 on Myriad

[taka]

When I run deeplab_mvn2 on NCS2, I get the following error.

In Openvino2020.2 I get an error. But with Openvino 2019 R3.1 it can run successfully

--------------------------------------------------

segmentation_demo.exe -d MYRIAD -m deeplabv3_mnv2.xml -i test.mp4

[ INFO ] InferenceEngine: 00007FFB3F734020
[ INFO ] Parsing input parameters
[ INFO ] Device info
        MYRIAD
        myriadPlugin version ......... 2.1
        Build ........... 42025
E: [global] [         0] [] XLinkResetRemote:249   can't wait dispatcherClosedSem


[ ERROR ] ArgMax/Squeeze of type Reshape: Stage node ArgMax/Squeeze (Reshape) types check error: output #0 has type FP16, but one of [S32] is expected

------------------------------------------------------

It occurs on Linux, Windows, HDDL, NCS2.

Where to get the model

https://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model_zoo.md

mobilenetv2_coco_voc_trainaug

I hope it will be fixed in the next version.

[RandallMan_B_Intel]

Hi taka,

Thanks for reaching out.

Could you provide the MO command used, and confirm if it is our segmentation demo?

Regards,

Randall B.

[Raza__Ghulam_Jilani]

Hi! I am having the same issue with a custom trained faster rcnnn model that i converted from pytorch and running on ncs2

"11101/Split of type Split: Check (actualType == expectedType) failed: Stage node 11101/Split types check error: output #1 has type FP16, but S32 is expected"

any ideas why this is happening? though i've set precision of all of my output nodes to FP32 in my c++ app, but this till keeps happening

[taka]

hi Randall

mo_tf.py ^

 --input_shape=[1,513,513,3]
 --input=1:mul_1
 --output=ArgMax
 --input_model=deeplabv3_mnv2_pascal_train_aug/frozen_inference_graph.pb
 --data_type=FP16

ran with segmanetation demo

# segmentation_demo.exe -d MYRIAD -m deeplabv3_mnv2.xml -i test.mp4

Best Regards,

[RandallMan_B_Intel]

Hi taka, Raza,

This error is due to an unsupported demo on the VPU devices, and may or may not work. The README, \deployment_tools\open_model_zoo\demos\segmentation_demo, for this demo says:

**NOTE**: On VPU devices (Intel® Movidius™ Neural Compute Stick, Intel® Neural Compute Stick 2, and Intel® Vision Accelerator Design with Intel® Movidius™ VPUs) this demo is not supported with any of the Model Downloader available topologies. Other models may produce unexpected results on these devices as well.

Take a look at the README here:

• https://docs.openvinotoolkit.org/latest/_demos_segmentation_demo_README.html
• https://docs.openvinotoolkit.org/latest/_demos_object_detection_demo_faster_rcnn_README.html

Regards,

Randall.

[taka]

hi Randall

I understand that it is not supported by NCS2.
However, since it worked on OpenVINO2019R3.1, I hope it will be supported someday!

Best Regards,