Where to use dropout in Resnets?

[roysubhankar]

Hi, Thanks for sharing the code for your paper.

Could you please tell me where I must use the dropout layers in a resnet50 for a classification task on 224x224 resolution images? Do you have some insights regarding where they must be used to obtain a good estimate of uncertainty?

Thanks.

[Hugo101]

Hi @roysubhankar , I have the same question. I am also wondering how to do this.
Hi @nayeemrizve , that would be really helpful if you could explain this.
I noticed that dropout is attached here in the shakeshakeblock:

ups/models/shakeshake.py

Line 226 in 89693c2

if self.dropout > 0:

Thank you so much!