output size

[Rinvay67]

how to deside the output size when chosing different network? or how to understand?
Thank you

[TontonTremblay]

The easiest way is to run forward with a random tensor, then look at its shape in the output. The original dope's output is 1/8. I have others in the new training script, I believe full is original input to original output. Then the others I am not sure, one is the original code base, and I believe I had one that did 1/2 as well.

[Rinvay67]

What I mean is why the output size is 400 or 50, how do you deside the output size or design the network? And if the output size is 50, the belief map's tensor is 50x50x9, the affinity_map's tensor is 50x50x16, right?
by the way, could you show the output image of belief map or affinity_map?
Thank you for your timely reply.