Pierre Fernandez

Hi and thanks for the quick answer. I've tried with dropout(0.3,0.3)+dropout(0.7,0.7)+cropout((0.84,0.84),(0.84,0.84))+cropout((0.55,0.55),(0.55,0.55))+crop((0.84,0.84),(0.84,0.84))+crop((0.55,0.55),(0.55,0.55))+blur(11,11)+blur(25,25)+jpeg() as noise, which should correspond to what is used in the paper, but I've obtained a BER of 0.1....

> I would like to ask you if there is possibility to modify the code to feed 3D images? if not, do you have a plan to extend the code...

> Also, as mentioned in the paper, concatenation allows to "retain [...] temporal information" compared to average pooling. > > > How many frames was picked for SSv2 task? >...

Closing as answered, thanks for your interest!

Hi, results from the paper are done using a single layer as linear classifier. What do you mean by "pretty low"? Would you be able to share your implementation and...

Thanks, I'll try to have a more thorough look at it in the following days. Some points that might cause the problem: - maybe try switching to SGD with different...

Thanks for keeping me updated on this @Batwho. Don't hesitate to re-open if you need anything else.