0

This question follows from my earlier question . I had converted this keras model to .pb. The I also successfully converted this .pb to intermediate representation of Open VINO. The model is available in the attachment of last answer of this post at intel developer forum.

If you got this model. Now you can load the model like this .

model_path = 'jester_freezed/IR/model/tf_model.xml'
from openvino.inference_engine import IENetwork, IECore
import os

def load_to_IE(model_xml, cpu_extension=None):
    #Load the Inference Engine API
    plugin = IECore()


    # Load IR files into their related class

    model_bin = os.path.splitext(model_xml)[0] + ".bin"

    net = IENetwork(model=model_xml, weights=model_bin)

    return net

net = load_to_IE(model_path)

Here is the full error.

RuntimeError: Check 'PartialShape::broadcast_merge_into(pshape, get_input_partial_shape(i), autob)' failed at /teamcity/work/scoring_engine_build/releases_2020_1/ngraph/src/ngraph/node.cpp:890: While validating node 'Add[Add_43](Convolution_41: float{1,40,12,12}, Constant_42: float{40}) -> (??)': Argument shapes are inconsistent.

My open vino version is 2020.1.

There are no resources available to debug this error. Please help. Thanks.

Deepanshu Yadav
  • 11
  • 6
  • hi, any progress with this problem , i am facing the same , i trained openvino text_detection with image size width 200px and height 120px, some how with default setting i can train my mode but trying that sample with text_detection c++ sample i got error, here is a question i asked on intel community please have a look https://community.intel.com/t5/Intel-Distribution-of-OpenVINO/any-documentation-for-training-text-detection/m-p/1193276 – Rawat Jul 18 '20 at 11:33
  • Hi how about [this](https://docs.openvinotoolkit.org/latest/omz_models_intel_text_recognition_0012_description_text_recognition_0012.html) may be. I gave up the this project and open vino , the Edge TPU from Google is much better choice , when it comes to performance , documentation and support. I have tried the coral board , the raspberry pi with Edge TPU and the NCS 2. I found the tflite paradigm much better to use and it performs much better. – Deepanshu Yadav Jul 18 '20 at 14:55

0 Answers0