Raspbian: bad_alloc in function 'initPlugin' on VGG19

[gomes__antonio]

Hello, 

I'm using RaspBerry + NCS2

I downloaded the VGG19 model through model_downloader and converted to IR.

when I try to run the model I have this  error:

File "teste1_vgg.py", line 45, in <module>

out = net.forward()
cv2.error: OpenCV(4.0.1-openvino) /home/jenkins/workspace/OpenCV/OpenVINO/build/opencv/modules/dnn/src/op_inf_engine.cpp:553: error: (-215:Assertion failed) Failed to initialize Inference Engine backend: std::bad_alloc in function 'initPlugin'

So I have an memory error in the function forward() but I don´t know why. 

 Can you help please?

Python Code: 

import cv2 as cv

net = cv.dnn.readNet('vgg19.xml', 'vgg19.bin') 
net.setPreferableTarget(cv.dnn.DNN_TARGET_MYRIAD)

frame_i = cv.imread('/home/pi/Desktop/modelos/classificacao/vgg19/car.bmp')
dim=(224,224)
frame = cv.resize(frame_i, dim)

blob = cv.dnn.blobFromImage(frame, size=(224, 224), ddepth=cv.CV_8U)
net.setInput(blob)
out = net.forward()

[Shubha_R_Intel]

Dear gomes, antonio,

Before trying your hand at coding from scratch, why not try one of our classification samples on the Model Optimizer Generated IR for VGG19 ?

You can try the classification_sample or classification_sample_async (both available in Python or C++) for example. 

Shubha

[gomes__antonio]

Hi Shubha,

Thanks for reply.

Yes, I already tried the classification sample and it worked. But not matters I switched the VGG19 for the Resnet and it work fine for what I want.

Now, I need a segmention model and I'm trying to convert the mask_rcnn_inception_resnet_v2_atrous_coco_2018_01_28 on link https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_convert_model_Convert_Model_From_TensorFlow.html but I have always a error:

I already tried the follow commands but none worked:

mo_tf.py --input_model D:\MYRIAD\mask_rcnn_inception_resnet_v2_atrous_coco_2018_01_28\frozen_inference_graph.pb

or

mo_tf.py --input_model D:\MYRIAD\mask_rcnn_inception_resnet_v2_atrous_coco_2018_01_28\frozen_inference_graph.pb --input_shape (1,640,425,3) --data_type=FP16

or

mo_tf.py --input_meta_graph D:\MYRIAD\mask_rcnn_inception_resnet_v2_atrous_coco_2018_01_28\model.ckpt.meta --input_shape (1,640,425,3) --data_type=FP16

or

mo_tf.py --input_model D:\MYRIAD\mask_rcnn_inception_resnet_v2_atrous_coco_2018_01_28\frozen_inference_graph.pb  --data_type=FP16

Is there a command missing? Thanks

Antonio

[Shubha_R_Intel]

Dear gomes, antonio,

Your mo_tf.py commands are not correct. Please follow the instructions here for converting Tensorflow Object Detection APIs via Model Optimizer:

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

Thanks,

Shubha