About the yolov3-tiny-prn.cfg file

[chauthehan]

I'm confused with shortcut=-3 and shortcut=8 in cfg file. As i understood, shortcut=-3 means we apply partial residual to the layer (i-1)th and layer (i-3)th. But what about shortcut=8. Please help. Many thanks.

[WongKinYiu]

-3 is relative index, 8 is absolute index.

[chauthehan]

thank you for your answer. I have another problem. I'm trying to implement yolov3-tiny-prn in tensorflow 2.0 , but it's seem like i misunderstood the shortcut block. In your paper you introduced about 3-5 prn architectures, how can I know which one did you use in yolo-tiny-prn. This is my implement https://github.com/chauthehan/yolov3_tiny_prn/blob/7a533838ded9b0ce0541d5c3f532378f6253a41e/yolov3_tiny_prn.py#L71 Many thanks.

[WongKinYiu]

better to use channel size to decide what you want to do.

channel1 = int((outputs[i-1].shape[-1]))
channel2 = int((outputs[i+from_].shape[-1]))

if channel1 == channel2:
    ...
elif channel1 > channel2:
    ...
else:
    ...