Scaled YOLO v4 Support

[pinczakko]

Does ncnn support for Scaled YOLOv4 (https://alexeyab84.medium.com/scaled-yolo-v4-is-the-best-neural-network-for-object-detection-on-ms-coco-dataset-39dfa22fa982) has been tested? (Note: pytorch implementation https://github.com/WongKinYiu/ScaledYOLOv4 ; Darknet implementation is in the original YOLOv4 github: https://github.com/AlexeyAB/darknet )
After looking at the code in both of the repositories, it seems that it doesn't need special activation function (as YOLO layers used in YOLO v4 is already supported in ncnn), only the network configuration differ. I recall that YOLOv4 support in ncnn for cfg and weight conversion from Darknet format to ncnn format (*.param and *.bin) is already mature, so the procedure to use Scaled YOLOv4 should still be the same as the current YOLOv4, right?

[pinczakko]

I've just tested the darknet2ncnn tools against the yolov4-csp weight and cfg. The conversion tool doesn't work, it loops forever in the yolov4 configuration file loading function ( load_cfg() ). I'll have a look at the conversion code more closely.