Yolo V3 demo error

[Lance_A_Intel]

Hello,

I am trying to run the windows C++ object_detection_demo_yolov3_async.exe and am getting error: "Error reading network: in Layer detector/yolo-v3-tiny/pool2/MaxPool: trying to connect an edge to non existing output port: 2.1”. See attached for command and result.

I am using OpenVINO version 19R1 (2019.1.087) with instructions on the demo here: https://docs.openvinotoolkit.org/latest/_inference_engine_samples_object_detection_demo_yolov3_async_README.html  

I have been using the instructions to obtain and convert the model here: https://software.intel.com/en-us/articles/OpenVINO-Using-TensorFlow#converting-a-darknet-yolo-model

• I downloaded the yolo v3  zip file commit ed60b90 from the git site
• I tried to convert and use both normal and tiny versions of yolo
• Using the Model Optimizer as documented produced an error about input of [-1 416 416 3] so I added argument –input_shape [1, 416, 416, 3] 
• I have tried different input videos as well as the camera
• I have tried on CPU and GPU
• all produce same error

[Shubha_R_Intel]

Dearest Lance, 

Did you follow the steps of this document carefully ? Also please make sure you use Tensorflow 1.11 or 1.12.

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

Did you build your IR correctly ? Please make sure you follow the yolo v3 instructions since the OpenVino samples are built for yolo3 not yolo2.

python3 mo_tf.py

--input_model /path/to/yolo_v3.pb

--tensorflow_use_custom_operations_config $MO_ROOT/extensions/front/tf/yolo_v3.json

I have run both the Python and C++ version of object_detection_demo_yolov3_async on video input with success on OpenVino 2019 R1.

Thanks,

Shubha

[Lance_A_Intel]

Can you provide the IR or the converted model?