A big problem with convert yolov3 to openvino format

[BuuVo]

Hi every one,

Following https://docs.openvinotoolkit.org/latest/openvino_docs_MO_DG_prepare_model_convert_model_tf_specific_Convert_YOLO_From_Tensorflow.html artical, I converted yolov3 to openvino format, and then I used this source code and modify a little bit  to do predictions. However the object outputs are always Zero and I don't know what is wrong with my conversion's step.

Here is the yolo_v3.json:

[
  {
    "id": "TFYOLOV3",
    "match_kind": "general",
    "custom_attributes": {
      "classes": 14,
      "anchors": [10, 13, 16, 30, 33, 23, 30, 61, 62, 45, 59, 119, 116, 90, 156, 198, 373, 326],
      "coords": 4,
      "num": 9,
      "masks":[[0, 1, 2], [3, 4, 5], [6, 7, 8]],
      "entry_points": ["detector/yolo-v3/Reshape", "detector/yolo-v3/Reshape_4", "detector/yolo-v3/Reshape_8"]
    }
  }
]

If any one know about my problem please let me know.

Thank you very much and best regard,

Buu Vo.

[BuuVo]

The source code that I follow and modify to predictions is: https://github.com/PINTO0309/OpenVINO-YoloV3/blob/master/openvino_yolov3_test.py

[BuuVo]

After converting yolo.weights to tensorflow protobuf, I do predictions again and the results are worse than original testing with yolo-darknet.

I used the below command to convert yolo to protobuf.

$python convert_weights_pb.py --class_names yolo.names --data_format NHWC --weights_file yolov3_last.weights

So, may I make something wrong when converting from yolo.weights to protobuf?

Thanks.

[IntelSupport]

Hi Buu Vo,

Thanks for reaching out. We are currently investigating this and will get back to you with the information at the earliest.

Regards,

Aznie

[IntelSupport · Accepted solution]

Hi Buu Vo,

I believe the conversion is correct since you are following the OpenVINO documentation. I found that this is a known issue with the program. There is a bug in the program that the frame display position is shifted and giving a poor performance. I believe you already following the correction suggested in the repository but unfortunately, there is nothing we can suggest for now since this is due to the bug with the program itself.

Regards,

Aznie

[BuuVo]

Thank you for you response.

[IntelSupport]

Hi Buu Vo,

This thread will no longer be monitored since this issue has been resolved. If you need any additional information from Intel, please submit a new question.

Regards,

Aznie