filter is batter than layer?

[qiu931110]

in the code,
bn.append(m1.weight.data.abs() / (m1.weight.data.abs() + m2.weight.data.abs()))

i don't understand why the weight better than enotrpy？

[fxmeng]

The BN layer controls the output of the filter by y = kx + b, where k is the weight of the BN layer and b is the bias of the BN layer. When the weight of the BN layer is zero, regardless of the value of the filter, the feature map of the next layer is 0, and this filter becomes an invalid filter. Therefore, the BN layer parameters can be used to measure the quality of the filter. In this way, using filter level on VGG can get better results than layer level. But this is not always the case, the layer level is still better on resnet.