run interactive_face_detection_demo on Raspberry failed

[BBi]

hi, after compiling interactive_face_detection_demo successfully on Raspberry, i ran it as below but failed---

root@raspberrypi:~/omz_demos_build/armv7l/Release# ./interactive_face_detection_demo -m /home/pi/Downloads/models/person-detection-retail-0013.xml -i /home/pi/Downloads/video/Cars1.mp4 -d MYRIAD
InferenceEngine: 0xb6dd107c
[ INFO ] Parsing input parameters
[ INFO ] Reading input
[ WARN:0] global ../opencv/modules/videoio/src/cap_gstreamer.cpp (898) open OpenCV | GStreamer warning: unable to query duration of stream
[ WARN:0] global ../opencv/modules/videoio/src/cap_gstreamer.cpp (935) open OpenCV | GStreamer warning: Cannot query video position: status=1, value=0, duration=-1
[ INFO ] Loading device MYRIAD
    MYRIAD
    myriadPlugin version ......... 2.1
    Build ........... 37988

[ INFO ] Loading network files for Face Detection
[ INFO ] Batch size is set to 1
[ INFO ] Checking Face Detection network inputs
[ INFO ] Checking Face Detection network outputs
[ INFO ] Loading Face Detection model to the MYRIAD device
[ INFO ] Age/Gender DISABLED
[ INFO ] Head Pose DISABLED
[ INFO ] Emotions Recognition DISABLED
[ INFO ] Facial Landmarks DISABLED
[ INFO ] Start inference 
corrupted double-linked list
Aborted

pi@raspberrypi:~ $ lsb_release -a
No LSB modules are available.
Distributor ID:    Raspbian
Description:    Raspbian GNU/Linux 10 (buster)
Release:    10
Codename:    buster
pi@raspberrypi:~ $ uname -a
Linux raspberrypi 4.19.97-v7l+ #1294 SMP Thu Jan 30 13:21:14 GMT 2020 armv7l GNU/Linux
=========================================

Q: i searched internet for the reason & solution, but most of them refer to memory leak issue. How can i fix this?

[Iffa_Intel]

Greetings,

First and foremost, I assume that you already know,

In order to run -d MYRIAD you must have a working NCS plugged in to your raspberry pi.

The pi must detect the presence of this stick, otherwise it wont work.

You should be good to go if you followed carefully the official Openvino tutorial here:

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

You can select version accordingly on your top right.

Please check the condition of your stick and if possible provide me your screen capture if you still have the same issue.

Hope my answer helps!

Sincerely,

Iffa

[BBi]

Greetings,

        Much appreciated for your kindly reply!

        And sorry for reply to you late cuz I was too busy.

As a reply, I have the working ncs2 at my hand and run benchmark / object_detection_sample_ssd with -d MYRIAD successfully, one screen capture like below---

BTW, openvino version I tried on rpi is 2020.1, but I think this issue isnot much related to openvino version and I can try the lates openvino in the near future.

I will study it and if any question I will feedback to you.

Again, thanks for your advice

[BBi]

attached picture 

[JAVIERJOSE_A_Intel]

Hi Bi, Barret,

As per this documentation the person-detection-retail-0013 model is not supported by the Interactive Face Detection C++ Demo. Interactive face detection demo only works with the following pre-trained models:

• face-detection-adas-0001
• age-gender-recognition-retail-0013
• head-pose-estimation-adas-0001
• emotions-recognition-retail-0003
• facial-landmarks-35-adas-0002

If you would like to use the person-detection-retail-0013 model with our samples/demos, use the Pedestrian Tracker C++ Demo.

NOTE: On VPU devices (Intel® Movidius™ Neural Compute Stick, Intel® Neural Compute Stick 2, and Intel® Vision Accelerator Design with Intel® Movidius™ VPUs) the Pedestrian Tracker C++ Demo has been tested on the following Model Downloader available topologies: >* person-detection-retail-0013 >* person-reidentification-retail-0031. Other models may produce unexpected results on these devices.

Regards,

Javier A.

[BBi]

Hi Javier,

          thanks for your kind reminder. i should use face-detection-adas-0001.xml matched this demo. again, the result is as below:

for benchmark test with ncs2---

[Step 10/11] Measuring performance (Start inference asyncronously, 4 inference requests, limits: 60000 ms duration)

[Step 11/11] Dumping statistics report
Count:      16836 iterations
Duration:   60018.33 ms
Latency:    14.14 ms
Throughput: 280.51 FPS
Peak Virtual Memory (VmPeak) Size, kBytes: 94736
Peak Resident Memory (VmHWM) Size, kBytes:  23036
pi@raspberrypi:~/build/armv7l/Release $ ./benchmark_app -m /home/pi/Downloads/models/squeezenet1.1/FP16/squeezenet1.1.xml -i /home/pi/Downloads/pic/car.png -d MYRIAD
 

for interactive_face_detection_demo test with ncs2--- using face-detection-adas-0001.xml

pi@raspberrypi:~/omz_demos_build/armv7l/Release $ ./interactive_face_detection_demo -m /home/pi/Downloads/face-detection-adas-0001.xml -i /dev/video0 -d MYRIAD
InferenceEngine: 0xb6e9807c
[ INFO ] Parsing input parameters
[ INFO ] Reading input
[ WARN:0] global ../opencv/modules/videoio/src/cap_gstreamer.cpp (1759) handleMessage OpenCV | GStreamer warning: Embedded video playback halted; module source reported: Could not read from resource.
[ WARN:0] global ../opencv/modules/videoio/src/cap_gstreamer.cpp (888) open OpenCV | GStreamer warning: unable to start pipeline
[ WARN:0] global ../opencv/modules/videoio/src/cap_gstreamer.cpp (480) isPipelinePlaying OpenCV | GStreamer warning: GStreamer: pipeline have not been created
[ INFO ] Loading device MYRIAD
    MYRIAD
    myriadPlugin version ......... 2.1
    Build ........... 37988

[ INFO ] Loading network files for Face Detection
[ ERROR ] Check 'axis < static_cast<size_t>(input_rank)' failed at /teamcity/work/scoring_engine_build/releases_2020_1/ngraph/src/ngraph/op/gather.cpp:140:
While validating node 'Gather[Gather_852](patternLabel_848: float{10,20,30}, patternLabel_849: int64_t{5}, patternLabel_851: int64_t{1}) -> (??)':
The axis must => 0 and <= input_rank (axis: 4294967295).

pi@raspberrypi:~/omz_demos_build/armv7l/Release $ 
 

i still remember this interactive_face_detection_demo can be run correctly last year when i tested the same command. so, is there anythin wrong with the GStreamer? or other reason caused this issue? 

[JAVIERJOSE_A_Intel]

Hi Barret,

The "check axis" error is caused by an incompatibility issue between the OpenVINO™ Toolkit 2020.1 version and the IR v10 format files.

As a workaround, you can try with one of the following options to solve this error:

Option 1:

Generate the IRv7 format files using the Model Optimizer from the full installation of the OpenVINO™ toolkit with the following parameter: --generate_deprecated_IR_V7.

Option 2:

Use the previous pre-trained models from the 2019 R3 open model zoo.

Option 3:

Install the previous version of the OpenVINO™ toolkit (2019 R3.1) on a Windows, Linux, or Mac OS system. Convert the model there will generate the IR v7 format.

Note: This bug has been fixed in the Intel® Distribution of OpenVINO™ toolkit v2020.2 for Raspbian OS. Update software to the latest version: https://download.01.org/opencv/2020/openvinotoolkit/

Regards,

Javier A.

[Iffa_Intel]

Hi,

From what my colleagues found previously,

The "axis" error in OpenVINO™ toolkit version (2020.1) probably caused by the new IR v10 format..

Try using the pre-trained model from the previous release:

wget https://download.01.org/opencv/2019/open_model_zoo/R3/20190905_163000_mo...

https://download.01.org/opencv/2019/open_model_zoo/R3/20190905_163000_mo.

If you are using the full installation of OpenVINO toolkit to optimize your own model, Please use this command  " --generate_deprecated_IR_V7 " if  full installation of OpenVINO toolkit was used for model optimization

This is  to flag  the model optimizer command to generate the previous version of IR format.

Hope this helps!. :)

Sincerely,

Iffa

[BBi]

Dear Javier & Iffa,

        thanks for your kindly reply.

i tried option 2 and ran  ./interactive_face_detection_demo -m /media/pi/0A09-8700/Rasberry/2019-R3-open-model-zoo/vehicle-detection-adas-0002.xml -i /home/pi/Downloads/video/Cars1.mp4 -d MYRIAD this same demo but sometimes ok(as the attached), sometimes ng(malloc():unsorted double linked list corrupted).

Myabe 2019R3.1 is better.

[Iffa_Intel]

Hi,

Any chance that you try this on Openvino 2020.2? The version 2020.1 does known to have certain errors with Raspbian/Raspberry Pi.

I did not have the chance to try it out yet but the community here https://github.com/openvinotoolkit/openvino/issues/411 mentioned they could run and their particular problems are solved.

Perhaps this would works for you too.

Thanks!

Sincerely,

Iffa