Yolo custom model to IR

[FSern]

Hello,

I want to convert a model which is based on yolov3 tiny into IR to use it with a movidius. its an ocr for license plates

I followed the guide provided in the docs with the pre-trained weights of the example and it worked,

however the model I want (the custom yolo ocr) gives me an error trying when I try to convert it to tf. the input shape on the cfg is 80 and 240 and the script provided receives squared shapes.

Is there something I can do to convert this custom yolo model into IR to be able to use it with openvino? or Am I doing something wrong?

Thanks.

[SuryaPSC_Intel]

Hi Felipe,

You can try changing the input placeholder code for converting weights to pb as per your model input shape. Also, you may have to make changes to the yolo_v3_tiny.json as per your custom model.

Best Regards,

Surya

[FSern]

Hi, Thanks for answering!

Turns out the model was based on yolo v2 tiny not v3, so I downloaded darkflow and was able to convert it to tensorflow. Tested it and looks fine.

However I have a new problem: I converted the .pb to openvino and everything looked fine in the conversion, but when I try to test the model doesn'y detect right.

I used this sample to test:

it https://github.com/opencv/open_model_zoo/blob/master/demos/python_demos/object_detection_demo_yolov3_async/object_detection_demo_yolov3_async.py

I though maybe the sample doesn't work with yolov2 but haven't been able to find a yolov2 openvino sample

I'm a bit lost, I dont know if its the sample or I converted wrong the model from tf to openvino

Thanks

[Vladimir_Dudnik]

Hi Felipe,

OMZ python object_detection_demo_yolov3_async has been tested with yolo models available in Open Model Zoo. They are:

yolo-v2-ava-0001

yolo-v2-ava-sparse-35-0001

yolo-v2-ava-sparse-70-0001

yolo-v2-tiny-ava-0001

yolo-v2-tiny-ava-sparse-30-0001

yolo-v2-tiny-ava-sparse-60-0001

yolo-v1-tiny-tf

yolo-v2-tf

yolo-v2-tiny-tf

yolo-v3-tf

It might be your model has some specific or difference in output?

Regards,
  Vladimir
 

[Alim_S_]

why, then?

object_detection_demo_yolov3_async -i cam -m C:\openvino\data\FP32\yolo-v2-tiny-ava-0001.xml

[ INFO ] Device info:
CPU
MKLDNNPlugin version ......... 2.1
Build ........... 2020.4.0-359-21e092122f4-releases/2020/4
[ INFO ] Loading network files
[ INFO ] Checking that the inputs are as the demo expects
[ INFO ] Checking that the outputs are as the demo expects
[ ERROR ] Size of dims(2) and format(NCHW) are inconsistent.