search

HobbitLong / RepDistiller

[ICLR 2020] Contrastive Representation Distillation (CRD), and benchmark of recent knowledge distillation methods
BSD 2-Clause "Simplified" License
2.11k stars 389 forks source link

resnet structure seems to be a bit wrong #46

Open surprisedong opened 2 years ago

surprisedong commented 2 years ago

resnet use 7x7conv and maxpool In the beginning，but this rep uses 3x3 conv and no maxpool，is there any reason for doing this?

surprisedong commented 2 years ago

the inputsize of cifar100 is too small to use 7X7 conv, i got it

surprisedong commented 2 years ago

is there any similar modify in other model structure?

cj-mclaughlin commented 2 years ago

@surprisedong If you review section 4.2 of the original ResNet paper (https://arxiv.org/pdf/1512.03385.pdf), you will see that for CIFAR the authors note all the alterations to the standard architecture that they use.