error on segmentation on NCS2

[tuaf__izack]

Hi,

I have a few questions regarding the abilities of Image Segmentation on the NCS2 stick.

I have successfully ran the "semantic-segmentation-adas-0001" model you provide in the model zoo:

Model description page: https://docs.openvinotoolkit.org/latest/_semantic_segmentation_adas_0001_description_semantic_segmentation_adas_0001.html

Model files link:

https://download.01.org/opencv/2019/open_model_zoo/R1/models_bin/semantic-segmentation-adas-0001/FP16/

About this model - "semantic-segmentation-adas-0001", can you provide details or links to the original pytorch model?

I would like to train a custom model based on this one (as it's actually working on the NCS2 comparing to other mask-rcnn models I’v tried)  and would be very helpful to know what layers where used or a link to the repository of the original model.

When I tried to run the instance-segmentation-security both 0049 and 0033 I get the following errors:

(I have tried all versions of the FP16 of both models in the repository, 2018 R3,R4,R5 & 2019 R1)

Error I get using the “instance-segmentation-security-0049” models:

python3 main.py -i cars.jpg -m instance-segmentatio

n-security-0049.xml --labels coco_labels.txt -d MYRIAD

[ INFO ] Initializing plugin for MYRIAD device...

[ INFO ] Reading IR...

[ INFO ] Loading IR to the plugin...

Traceback (most recent call last):

 File "main.py", line 269, in <module>

   sys.exit(main() or 0)

 File "main.py", line 154, in main

   exec_net = plugin.load(network=net, num_requests=2)

 File "ie_api.pyx", line 395, in openvino.inference_engine.ie_api.IEPlugin.load

 File "ie_api.pyx", line 406, in openvino.inference_engine.ie_api.IEPlugin.load

RuntimeError: [VPU] Cannot convert layer "554" due to unsupported layer type "ExperimentalDetectronPriorGridGenerator"

Error I get using the “instance-segmentation-security-0033” models:

python3 main.py -i cars.jpg -m instance-segmentatio

n-security-0033.xml --labels coco_labels.txt -d MYRIAD

[ INFO ] Initializing plugin for MYRIAD device...

[ INFO ] Reading IR...

[ INFO ] Loading IR to the plugin...

Traceback (most recent call last):

 File "main.py", line 269, in <module>

   sys.exit(main() or 0)

 File "main.py", line 154, in main

   exec_net = plugin.load(network=net, num_requests=2)

 File "ie_api.pyx", line 395, in openvino.inference_engine.ie_api.IEPlugin.load

 File "ie_api.pyx", line 406, in openvino.inference_engine.ie_api.IEPlugin.load

RuntimeError: [VPU] Cannot convert layer "1510" due to unsupported layer type "ExperimentalDetectronPriorGridGenerator"

I have also tried from a tutorial by the awesome PyImageSearch an ENET model:

https://www.pyimagesearch.com/2018/09/03/semantic-segmentation-with-opencv-and-deep-learning/

It works on CPU but not on NCS2 VPU with the following error:

[INFO] loading model...

E: [xLink] [    392966] XLinkOpenStream:1174 Got wrong package from device, error code = X_LINK_OUT_OF_MEMORY

Segmentation fault (core dumped)

All tests ran on an Ubuntu 16.04.6 VM with the NCS2 stick.

Is there an upcoming update to OpenVINO to support those models and similar on the NCS2?

Thank you very much,

Izack.

[Lazarescu__Bogdan]

Hello! 

I am interested in performing the same task. I need a image object segmentation algorithm. Unfortunately  "semantic-segmentation-adas-0001" available in the zoo models runs a bit too slow for my needs. Did you find any solution?

[Shubha_R_Intel]

Dear tuaf, izack and Lazarescu, Bogdan,

You can use model_downloader, i.e. python downloader.py --all to download semantic_segmentation models OpenVino supports (these are different from the ones in the model zoo located here ). You can also run inference on those models by using OpenVino's C++ and Python segmentation demos. More info here:

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

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

If you're building your own semantic segmentation model from scratch for MYRIAD, it's indeed possible that some layers may not be supported. Below is documentation on supported MYRIAD layers:

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

Hope it helps. And thanks for using OpenVino !

Shubha