-
你好,看到你的readme介绍,你实现了规整剪枝、正常剪枝和分组卷积结构剪枝,我理解规整和正常剪枝是实现了slimming论文里的剪枝方法是吗,那分组卷积剪枝具体实现的是Rethinking the Value of Network Pruning论文里提到的哪个方法呢?
-
大牛,我想问一个问题,我觉的论文《Learning Efficient Convolutional Networks Through Network Slimming》中给出的损失函数是针对需要剪枝的BN层的,而网络的最后层的损失函数还是经典的yolov3的损失函数,可以这样理解吗?根据代码的意思,最后的loss依然是经典的yolov3的损失函数值,没有加入L1正则的损失值
期待您的回复。十…
-
您好,请问为什么我们需要channel selection layer来辅助ResNet和DenseNet的剪枝呀?
我看代码,自己的理解是对于ResNet和DenseNet在BN层后面添加了channel selection layer,然后进行训练。在模型裁剪的时候,channel selection layer的值全置0,然后将需要保留的赋值为1.
```
# We nee…
-
大牛,我想问一个问题,我觉的论文《Learning Efficient Convolutional Networks Through Network Slimming》中给出的损失函数是针对需要剪枝的BN层的,而网络的最后层的损失函数还是经典的yolov3的损失函数,可以这样理解吗?根据代码的意思,最后的loss依然是经典的yolov3的损失函数值,没有加入L1正则的损失值
期待您的回复。十…
-
原论文中的表6中的memory指的是什么?我训练出的vgg16的权重文件只有118MB,可表6中的baseline的memory为697MB,这两个不是一个东西吗
-
你好,感谢您的分享。
請問可以說明一下關於L1 scales regularization 方法具體需要修改源碼那些地方嗎?
-
大神,你好。我想问一个问题,我觉的论文《Learning Efficient Convolutional Networks Through Network Slimming》中给出的损失函数是针对需要剪枝的BN层的,而网络的最后层的损失函数还是经典的yolov3的损失函数,可以这样理解吗?根据代码的意思,最后的loss依然是经典的yolov3的损失函数值,没有加入L1正则的损失值。如果在网络的最后…
-
### Describe the question you meet
Can autoslim be used on mmsegmentation models like [pspnet](https://github.com/open-mmlab/mmsegmentation/tree/master/configs/pspnet) or [segformer](https://github.c…
-
Dear @liuzhuang13,
I guess we should prune some channel of subsequent conv layer' kernels after pruning current layer. Am I right?
So I can not figure out how to slim residual block using your m…
-
Hi,
There is no retrain.yml available in the configs folder. I am wondering how to retrain the searched models for cityscapes.
Also, when I train the search model with the seg_cityscapes.yml, th…