如何更改 Pytorch 预训练模块中的激活层？

Question

如何更改 Pytorch 预训练网络的激活层？ 这是我的代码：

print("All modules")
for child in net.children():
    if isinstance(child,nn.ReLU) or isinstance(child,nn.SELU):
        print(child)

print('Before changing activation')
for child in net.children():
    if isinstance(child,nn.ReLU) or isinstance(child,nn.SELU):
        print(child)
        child=nn.SELU()
        print(child)
print('after changing activation')
for child in net.children():
    if isinstance(child,nn.ReLU) or isinstance(child,nn.SELU):
        print(child)

这是我的输出：

All modules
ReLU(inplace=True)
Before changing activation
ReLU(inplace=True)
SELU()
after changing activation
ReLU(inplace=True)

Answer 1

._modules 为我解决了问题。

for name,child in net.named_children():
    if isinstance(child,nn.ReLU) or isinstance(child,nn.SELU):
        net._modules['relu'] = nn.SELU()

Answer 2

我假设您使用模块接口nn.ReLU 来创建激活层，而不是使用功能接口F.relu。如果是这样，setattr 对我有用。

import torch
import torch.nn as nn

# This function will recursively replace all relu module to selu module. 
def replace_relu_to_selu(model):
    for child_name, child in model.named_children():
        if isinstance(child, nn.ReLU):
            setattr(model, child_name, nn.SELU())
        else:
            replace_relu_to_selu(child)

########## A toy example ##########
net = nn.Sequential(
            nn.Conv2d(3, 32, kernel_size=3, stride=1),
            nn.ReLU(inplace=True),
            nn.Conv2d(3, 32, kernel_size=3, stride=1),
            nn.ReLU(inplace=True)
          )

########## Test ##########
print('Before changing activation')
for child in net.children():
    if isinstance(child,nn.ReLU) or isinstance(child,nn.SELU):
        print(child)
# Before changing activation
# ReLU(inplace=True)
# ReLU(inplace=True)


print('after changing activation')
for child in net.children():
    if isinstance(child,nn.ReLU) or isinstance(child,nn.SELU):
        print(child)
# after changing activation
# SELU()
# SELU(

Answer 3

这是替换任何层的通用函数

def replace_layers(model, old, new):
    for n, module in model.named_children():
        if len(list(module.children())) > 0:
            ## compound module, go inside it
            replace_layers(module, old, new)
            
        if isinstance(module, old):
            ## simple module
            setattr(model, n, new)

replace_layer(model, nn.ReLU, nn.ReLU6())

我为此苦苦挣扎了几天。所以，我做了一些挖掘并写了一个kaggle notebook 解释如何在 pytorch 中访问不同类型的层/模块。

Answer 4

我将提供一个适用于任何层的更通用的解决方案（并避免其他问题，例如在循环遍历字典时修改字典或彼此内部存在递归 nn.modules 时）。

def replace_bn(module, name):
    '''
    Recursively put desired batch norm in nn.module module.

    set module = net to start code.
    '''
    # go through all attributes of module nn.module (e.g. network or layer) and put batch norms if present
    for attr_str in dir(module):
        target_attr = getattr(m, attr_str)
        if type(target_attr) == torch.nn.BatchNorm2d:
            print('replaced: ', name, attr_str)
            new_bn = torch.nn.BatchNorm2d(target_attr.num_features, target_attr.eps, target_attr.momentum, target_attr.affine,
                                          track_running_stats=False)
            setattr(module, attr_str, new_bn)

    # iterate through immediate child modules. Note, the recursion is done by our code no need to use named_modules()
    for name, immediate_child_module in module.named_children():
        replace_bn(immediate_child_module, name)

replace_bn(model, 'model')

关键是您需要递归地不断更改层（主要是因为有时您会遇到本身具有模块的属性）。我认为比上面更好的代码是添加另一个 if 语句（在批处理规范之后）检测是否必须递归，如果是，则递归。上面的方法首先改变了外层的批处理规范（即第一个循环），然后用另一个循环确保没有其他应该递归的对象被遗漏（然后递归）。

原帖：https://discuss.pytorch.org/t/how-to-modify-a-pretrained-model/60509/10

学分：https://discuss.pytorch.org/t/replacing-convs-modules-with-custom-convs-then-notimplementederror/17736/3?u=brando_miranda