I was recently trying to perform prediction on images with a pre-trained yolov3 model that I had already converted to keras with convert.py. The issue was that the path options such as --classes or --model where not being taken into account.

After further digging, it appears that the argparse and attribute names don't match. For instance with the model path:

• In yolo_video.py -> model.
• In yolo.py -> model_path.

I've found an easy fix by adding a destination tag to the argparse:

• model: '--model', type=str, dest='model_path',
• anchors: '--anchors', type=str, dest='anchors_path',
• classes: '--classes', type=str, dest='classes_path',

Do you know if I missed anything?