Converting yolov3

[pkhan10]

Hello
In the documentation of conversion for yolov3 following git was used

https://github.com/mystic123/tensorflow-yolo-v3

Unfortunately there's no instruction to train custom model in the above git. I used 
https://github.com/YunYang1994/tensorflow-yolov3 to train my model
Can you please help me to convert model to openvino format

[Iffa_Intel]

Greetings,

First and foremost, you need to identify your model is frozen or not. If it is already frozen, you can directly convert them

If not, There are three ways to store non-frozen TensorFlow models and load them to the Model Optimizer

1. Checkpoint
2. MetaGraph
3. SavedModel format of TensorFlow 1.x and 2.x versions

You may refer to this official documentation: https://docs.openvinotoolkit.org/latest/openvino_docs_MO_DG_prepare_model_convert_model_Convert_Model_From_TensorFlow.html

This guide is specific for yolov3 and others: https://docs.openvinotoolkit.org/latest/openvino_docs_MO_DG_prepare_model_convert_model_tf_specific_Convert_YOLO_From_Tensorflow.html

you may also refer to this video for detailed steps and results: https://www.youtube.com/watch?v=FaqVhvJ6-Uc

Sincerely,

Iffa

[pkhan10]

hello iffa
as clearly mentioned in documentation, that the conversion is only provided for https://github.com/mystic123/tensorflow-yolo-v3

While above repo did have conversion approach for the yolov3 original weights, but it doesn't have training pipeline.
I used https://github.com/YunYang1994/tensorflow-yolov3to train my custom model, but unable to convert model into openvino format
Can you help me convert the model through above pipeline, if not that can you suggest training repos that can be used to train and later on convert into openvino format

[Iffa_Intel]

Greetings,

I had tested the model that you mentioned & I also found the same problem which there is no compatible json file available to be used with this model.

Openvino actually does have readily available json file which you can download using downloader.py in the location openvino\deployment_tools\tools\model_downloader

And also in openvino/deployment_tools/model_optimizer/extensions/front/tf

I had tried with all the possible yolov3 json and it actually doesn't works and this can be expected as this model is never been validated as the mystic123 model.

This model might have a different structure. You may reder to this example on how to create a yolov3 model with the supported structure.

Sincerely,

Iffa

[Iffa_Intel]

Greetings,

Intel will no longer monitor this thread since we have provided a solution. If you need any additional information from Intel, please submit a new question. 

Sincerely,

Iffa