Convert darknet weights to tensorflow

[vqbang]

Hi @AlexeyAB, I'm trying to find a repo to convert the weights that I'd trained using your repo, with a custom config based on yolov3.cfg file. After some researches, these are the repos that I've tried:

• https://github.com/mystic123/tensorflow-yolo-v3
• https://github.com/ultralytics/yolov3.git
• https://github.com/YunYang1994/TensorFlow2.0-Examples.git
• https://github.com/zzh8829/yolov3-tf2.git
• https://github.com/hunglc007/tensorflow-yolov4-tflite
• https://github.com/pythonlessons/TensorFlow-2.x-YOLOv3.git

But It seems like none of them have the same result as yours. The most accurate one is mystic123, but still have some slight differences (about 2-3% accuracy) per prediction. Have you tried to convert a successful one to tensorflow with no changes in result? Have a good day my friend :D

[GotG]

I have the same problem. My original yolo model has more detections (on the same image) in darknet than in the converted tensorflow model using mystic123. It would be nice to obtain (more) similar results on the same image with both original and converted model.