About the implementation of stochastic depth

[mai556]

Hi！
In the original paper of stochastic depth, The formula is expressed as

Why mention it be a bug? And fix it as

[Zhou2019]

hi, have you found out the reason?

[Zhou2019]

hi, I have found the reason in the paper "A Survey on Dropout Methods and Experimental Verification in Recommendation“, which said " To keep training and testing conditions consistent, all the weights need to be multiplied by p during testing, i.e., take W(l)test = pW(l); or it multiplies all outputs by 1/p during training so that the expected output is consistent with testing time. "