Converting YOLOv4 model implemented in Darknet framework to TensorFlow framework

[vk_007]

I want to use my trained real time object detection YOLOv4 model in openvino to convert it to the Intermediate Representation (IR). But the trained YOLOv4 model is implemented in Darknet framework and in the openvino we have to first convert our YOLOv4 model to TensorFlow framework in order to use our trained model.

So, is there any methods where we can convert our YOLOv4 model to TensorFlow?

[IntelSupport]

Hi vk_007,

Thanks for reaching out.

There is no method from OpenVINO to convert the YOLOv4 model into the TensorFlow model. However, there is a method of Converting YOLO* Models to the Intermediate Representation (IR) that you can try. But, this method was only validated for YOLOv1, YOLOv2, and YOLOv3 only.

Meanwhile, for your information, the yolo-v4-tf model is available in OpenVINO object detection model.

Regards,

Aznie

[IntelSupport]

Hi vk_007,

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

Regards,

Aznie

[Aznie_Intel]

Hi vk_007,

For your information, the conversion steps of YOLOv4  model is now available in OpenVINO 2021.4 toolkit. 

Regards,

Aznie 

[Pranathi]

Hi! When I try to use the instructions in the link posted to convert my custom dataset .weights file for tiny yolo v4 model to .pb file, I get no matching filenames error for the file name I have

yolov4-tiny-obj_best.weights

What are the filename patterns it accepts? Also, looks like this documentation is only for yolo models and not tiny yolo models, could you please point me to that? Thanks!