why add bn_value in imagenet_resnet_small.py

he-y / soft-filter-pruning

Soft Filter Pruning for Accelerating Deep Convolutional Neural Networks

https://arxiv.org/abs/1808.06866

376 stars 74 forks source link

Open Emily0219 opened 4 years ago

Emily0219 commented 4 years ago

你好，这里的bn_value不就是bn3.weight吗？为什么residual还要再加一次呢？ `

    residual += self.bn_value.cuda()
    residual.index_add_(1, self.index.cuda(), out)
    residual = self.relu(residual)

Emily0219 commented 4 years ago

看您之前的解释，是零输入对应的bn层输出，那为什么不减去bn_value呢？是因为未剪枝前bn就有输出，所以剪完枝之后要加上吗？

zhangxin-xd commented 2 years ago

Hi， I also don't understand this operation. Do you have any answers? @Emily0219

zhangxin-xd commented 2 years ago

应该是这样的，是因为bn_value是对应剪掉位置的bn值，self.bn3的大小则和conv3保持一致（只有remaining 的filter对应的值）， bn_value相当于0输入时，剪掉部分的对应输出，所以需要加上