BeSpontaneous/FFN-pytorch

Fig 3 in paper

Closed this issue · 2 comments

I have some doubts about Fig 3 in paper. How is the layer index defined in resnet50 and how to calculate the value of Mean and Variance. Thank you.

Hi,

(1) As illustrated in the caption of Fig. 3, the statistics are calculated from the fourth stage of ResNet-50 and ResNet-50 has 3 bottlenecks at the 4th stage which consists of 9 layers.
(2) Mean and Var can be obtained from the pre-trained model in PyTorch such as:
mean = model.module.base_model.layer4[0].bn1.running_mean
var = model.module.base_model.layer4[0].bn1.running_var

thanks a lot for your reply.