mask_rcnn_resnet101_atrous_coco_2018_01_28 Fail in MYRIAD(NCS2))

[BSung8]

Dear Sir,

I converted mask_rcnn_resnet101_atrous_coco_2018_01_28 model and  inference are functioning properly in CPU.

But it's fail in MYRIAD with FP16 format.

error message as below,

InferenceEngine:

    API version ............ 1.6

    Build .................. 23780

[ INFO ] Parsing input parameters

[ INFO ] Files were added: 1

[ INFO ]   C:\Users\bryans\tensorflow_workspace\tensorflow\models\research\object_detection\test_images\image2.jpg

[ INFO ] Loading plugin

    API version ............ 1.6

    Build .................. 23780

    Description ....... myriadPlugin

[ INFO ] Loading network files

[ INFO ] Preparing input blobs

[ INFO ] Network batch size is 1

[ INFO ] Prepare image C:\Users\bryans\tensorflow_workspace\tensorflow\models\research\object_detection\test_images\image2.jpg

[ WARNING ] Image is resized from (1352, 900) to (800, 800)

[ INFO ] Preparing output blobs

[ INFO ] Loading model to the plugin

[ ERROR ] [VPU] Can't satisfy data location requirements for stage SecondStageBoxPredictor_1/ResizeBilinear@batch=1/100

My OpenVINO version : openvino_2019.1.148

Thanks!!

[Shubha_R_Intel]

Dear Sung, Bryan,

Are you running the C++ mask_rcnn_demo ? If not, which specific sample are you running ? Thanks for updating your OpenVino to 2019R1.1 which is the latest.

Shubha

[BSung8]

Dear Shubha,

Yes, I'm running mask_rcnn_demo. Windows version, path as below

C:\Program Files (x86)\IntelSWTools\openvino_2019.1.148\deployment_tools\inference_engine\samples\mask_rcnn_demo

model download link as below,

http://download.tensorflow.org/models/object_detection/mask_rcnn_resnet101_atrous_coco_2018_01_28.tar.gz

Thanks!!

[Shubha_R_Intel]

Dear Sung, Bryan,

I'm afraid you found a bug which I was able to reproduce on an even later build than 2019R1.1. I have filed a bug on your behalf. Thank you for your patience !

Stay tuned for OpenVino 2019R2.

Thanks,

Shubha

[Shubha_R_Intel]

Dear Sung, Bryan,

I wanted to give you an update. Basically you ran into a memory limitation on the MYRIAD.  We are still investigating this.

That said, it could have certainly been handled better. We could have reported an "Out of Memory" exception or something along those lines.

Thanks for your patience !

Shubha