Is scaled-YOLOv4 going to be supported?

[hhk7734]

Supports yolov4-csp and yolov4x-mish without training part.

Some cfg options are not yet supported(almost training part options). So, download the config file in this repo. https://github.com/hhk7734/tensorflow-yolov4/tree/master/config

from yolov4.tf import YOLOv4, save_as_tflite

yolo = YOLOv4()

yolo.config.parse_names("coco.names")
yolo.config.parse_cfg("yolov4-csp.cfg")

yolo.make_model()
yolo.load_weights("yolov4-csp.weights", weights_type="yolo")
yolo.summary(summary_type="yolo")
yolo.summary()

yolo.inference(media_path="kite.jpg")

yolo.inference(media_path="road.mp4", is_image=False)

yolo.inference(
    "/dev/video0",
    is_image=False,
    cv_apiPreference=cv2.CAP_V4L2,
    cv_frame_size=(640, 480),
    cv_fourcc="YUYV",
)

[4ku]

@hhk7734 Thank you) And what about scaled tiny model?

[hhk7734]

Did you mean config/yolov4-tiny-relu-new_coords.cfg? I'm testing it because I thought it might show a bit faster speed on the EdgeTpu. I'm busy these days, so I paused for a while.

[4ku]

I mean this. But I don't know, may be it's the same.

[hhk7734]

It seems to be just yolov4 tiny.

[4ku]

May be, but configs differs a little

[hhk7734]

The options that differ are probably training-related options.

[4ku]

ok, thank you