深度学习pytorch分层、分参数设置不同的学习率等

最新推荐文章于 2022-11-13 12:44:00 发布

dadaHaHa1234

最新推荐文章于 2022-11-13 12:44:00 发布

阅读量5.5k

点赞数 3

本文链接：https://blog.csdn.net/qq_32425195/article/details/85240854

版权

对于bias参数和weight参数，设置不同的学习率

{'params': get_parameters(model, bias=False)},
{'params': get_parameters(model, bias=True),
'lr': args.lr * 2, 'weight_decay': 0},

对于不同的层，做不同的处理

for m in model.modules():

if xxx

else if xxx

def get_parameters(model, bias=False):
    import torch.nn as nn
    modules_skipped = (
        nn.ReLU,
        nn.MaxPool2d,
        nn.Dropout2d,
        nn.Sequential,
        torchfcn.models.FCN32s,
        torchfcn.models.FCN16s,
        torchfcn.models.FCN8s,
    )
    for m in model.modules():
        if isinstance(m, nn.Conv2d):
            if bias:
                yield m.bias
            else:
                yield m.weight
        elif isinstance(m, nn.ConvTranspose2d):
            # weight is frozen because it is just a bilinear upsampling
            if bias:
                assert m.bias is None
        elif isinstance(m, modules_skipped):
            continue
        else:
            raise ValueError('Unexpected module: %s' % str(m))


    optim = torch.optim.SGD(
        [
            {'params': get_parameters(model, bias=False)},
            {'params': get_parameters(model, bias=True),
             'lr': args.lr * 2, 'weight_decay': 0},
        ],
        lr=args.lr,
        momentum=args.momentum,
        weight_decay=args.weight_decay)

关键点解析：

点1：首先，

for m in model.modules()中，m依次是什么

以自定义的网络FCN32s为例，m是FCN32s网络module和FCN32s的init函数（只包括init（）函数）中所有定义的net module

代码：

自定义的FCN32s网络

class FCN32s(nn.Module):    
    def __init__(self, n_class=21):
        super(FCN32s, self).__init__()
        #将每个层都定义为FCN32s类的参数
        # conv1
        self.conv1_1 = nn.Conv2d(3, 64, 3, padding=100)
        self.relu1_1 = nn.ReLU(inplace=True)
        self.conv1_2 = nn.Conv2d(64, 64, 3, padding=1)
        self.relu1_2 = nn.ReLU(inplace=True)
        self.pool1 = nn.MaxPool2d(2, stride=2, ceil_mode=True)  # 1/2

        # conv2
        self.conv2_1 = nn.Conv2d(64, 128, 3, padding=1)
        self.relu2_1 = nn.ReLU(inplace=True)
        self.conv2_2 = nn.Conv2d(128, 128, 3, padding=1)
        self.relu2_2 = nn.ReLU(inplace=True)
        self.pool2 = nn.MaxPool2d(2, stride=2, ceil_mode=True)  # 1/4

        # conv3
        self.conv3_1 = nn.Conv2d(128, 256, 3, padding=1)
        self.relu3_1 = nn.ReLU(inplace=True)
        self.conv3_2 = nn.Conv2d(256, 256, 3, padding=1)
        self.relu3_2 = nn.ReLU(inplace=True)
        self.conv3_3 = nn.Conv2d(256, 256, 3, padding=1)
        self.relu3_3 = nn.ReLU(inplace=True)
        self.pool3 = nn.MaxPool2d(2, stride=2, ceil_mode=True)  # 1/8

        # conv4
        self.conv4_1 = nn.Conv2d(256, 512, 3, padding=1)
        self.relu4_1 = nn.ReLU(inplace=True)
        self.conv4_2 = nn.Conv2d(512, 512, 3, padding=1)
        self.relu4_2 = nn.ReLU(inplace=True)
        self.conv4_3 = nn.Conv2d(512, 512, 3, padding=1)
        self.relu4_3 = nn.ReLU(inplace=True)
        self.pool4 = nn.MaxPool2d(2, stride=2, ceil_mode=True)  # 1/16

        # conv5
        self.conv5_1 = nn.Conv2d(512, 512, 3, padding=1)
        self.relu5_1 = nn.ReLU(inplace=True)
        self.conv5_2 = nn.Conv2d(512, 512, 3, padding=1)
        self.relu5_2 = nn.ReLU(inplace=True)
        self.conv5_3 = nn.Conv2d(512, 512, 3, padding=1)
        self.relu5_3 = nn.ReLU(inplace=True)
        self.pool5 = nn.MaxPool2d(2, stride=2, ceil_mode=True)  # 1/32

        # fc6
        self.fc6 = nn.Conv2d(512, 4096, 7)
        self.relu6 = nn.ReLU(inplace=True)
        self.drop6 = nn.Dropout2d()

        # fc7
        self.fc7 = nn.Conv2d(4096, 4096, 1)
        self.relu7 = nn.ReLU(inplace=True)
        self.drop7 = nn.Dropout2d()


        #用于计算每个类的分值
        self.score_fr = nn.Conv2d(4096, n_class, 1)
        #ConvTranspose2d是Conv2d的逆操作，此处输入通道数和输出通道数都是n_class，上采样卷积核为64，步进为32.上采样反卷积的操作参考github
        self.upscore = nn.ConvTranspose2d(n_class, n_class, 64, stride=32, bias=False)

        self._initialize_weights()

net.modules（）打印的结果

FCN32s(
  (conv1_1): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(100, 100))
  (relu1_1): ReLU(inplace)
  (conv1_2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu1_2): ReLU(inplace)
  (pool1): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=True)
  (conv2_1): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu2_1): ReLU(inplace)
  (conv2_2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu2_2): ReLU(inplace)
  (pool2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=True)
  (conv3_1): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu3_1): ReLU(inplace)
  (conv3_2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu3_2): ReLU(inplace)
  (conv3_3): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu3_3): ReLU(inplace)
  (pool3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=True)
  (conv4_1): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu4_1): ReLU(inplace)
  (conv4_2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu4_2): ReLU(inplace)
  (conv4_3): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu4_3): ReLU(inplace)
  (pool4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=True)
  (conv5_1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu5_1): ReLU(inplace)
  (conv5_2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu5_2): ReLU(inplace)
  (conv5_3): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
  (relu5_3): ReLU(inplace)
  (pool5): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=True)
  (fc6): Conv2d(512, 4096, kernel_size=(7, 7), stride=(1, 1))
  (relu6): ReLU(inplace)
  (drop6): Dropout2d(p=0.5)
  (fc7): Conv2d(4096, 4096, kernel_size=(1, 1), stride=(1, 1))
  (relu7): ReLU(inplace)
  (drop7): Dropout2d(p=0.5)
  (score_fr): Conv2d(4096, 21, kernel_size=(1, 1), stride=(1, 1))
  (upscore): ConvTranspose2d(21, 21, kernel_size=(64, 64), stride=(32, 32), bias=False)
)
Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(100, 100))
ReLU(inplace)
Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU(inplace)
MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=True)
Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU(inplace)
Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU(inplace)
MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=True)
Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU(inplace)
Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU(inplace)
Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU(inplace)
MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=True)
Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU(inplace)
Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU(inplace)
Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU(inplace)
MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=True)
Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU(inplace)
Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU(inplace)
Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
ReLU(inplace)
MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=True)
Conv2d(512, 4096, kernel_size=(7, 7), stride=(1, 1))
ReLU(inplace)
Dropout2d(p=0.5)
Conv2d(4096, 4096, kernel_size=(1, 1), stride=(1, 1))
ReLU(inplace)
Dropout2d(p=0.5)
Conv2d(4096, 21, kernel_size=(1, 1), stride=(1, 1))
ConvTranspose2d(21, 21, kernel_size=(64, 64), stride=(32, 32), bias=False)

Process finished with exit code 0

这样，就可以通过for m in net.modules对FCN32s中不同的网络子模块做不同的处理。

每次遍历时，m的类型时nn.Modules神经网络，m.bias的类型是tensor

点2：

{'params': get_parameters(model, bias=False)}中，get_parameters返回的是某个神经网络的参数，是一个tensor。我猜测，当参数更新时，要更新到这个参数时，就会查询这个参数的更新方法，而这个参数的更新方法就是通过上面截图来指定的

dadaHaHa1234

关注

3
点赞
踩
4

收藏

觉得还不错? 一键收藏
3
评论
深度学习pytorch分层、分参数设置不同的学习率等

1.对于bias参数和weight参数，设置不同的学习率{'params': get_parameters(model, bias=False)}, {'params': get_parameters(model, bias=True), 'lr': args.lr * 2, 'weight_decay': 0},对于不同的层，做不同的...
复制链接

扫一扫