Tensorflow 张量流的Pytorch等价物

我正试图在Pytorch中跟踪此代码。我已经试了好几天了，但阅读张量流文档和PyTorch文档让我完全困惑

input_data = Input(shape=(256, 64, 1), name=‘input’)

inner = Conv2D(32, (3, 3), padding=‘same’, name=‘conv1’, kernel_initializer=‘he_normal’)(input_data)
inner = BatchNormalization()(inner)
inner = Activation(‘relu’)(inner)
inner = MaxPooling2D(pool_size=(2, 2), name=‘max1’)(inner)

inner = Conv2D(64, (3, 3), padding=‘same’, name=‘conv2’, kernel_initializer=‘he_normal’)(inner)
inner = BatchNormalization()(inner)
inner = Activation(‘relu’)(inner)
inner = MaxPooling2D(pool_size=(2, 2), name=‘max2’)(inner)
inner = Dropout(0.3)(inner)

inner = Conv2D(128, (3, 3), padding=‘same’, name=‘conv3’, kernel_initializer=‘he_normal’)(inner)
inner = BatchNormalization()(inner)
inner = Activation(‘relu’)(inner)
inner = MaxPooling2D(pool_size=(1, 2), name=‘max3’)(inner)
inner = Dropout(0.3)(inner)

CNN to RNN
inner = Reshape(target_shape=((64, 1024)), name=‘reshape’)(inner)
inner = Dense(64, activation=‘relu’, kernel_initializer=‘he_normal’, name=‘dense1’)(inner)

RNN
inner = Bidirectional(LSTM(256, return_sequences=True), name = ‘lstm1’)(inner)
inner = Bidirectional(LSTM(256, return_sequences=True), name = ‘lstm2’)(inner)

OUTPUT
inner = Dense(num_of_characters, kernel_initializer=‘he_normal’,name=‘dense2’)(inner)
y_pred = Activation(‘softmax’, name=‘softmax’)(inner)

model = Model(inputs=input_data, outputs=y_pred)

我试着在Pytorch中一点一点地跟踪它

class Net(nn.Module):
def init(self):
super(Net,self).init()
self.input_data = input_size
self.conv1 = nn.Conv2d(32,3,3)
self.conv2 = nn.Conv2d(64,3,3)
self.conv3 = nn.Conv2d(128,3,3)


self.dropout = nn.Dropout(0.3)
self.maxp = torch.nn.MaxPool2d((2,2))


#CNN to RNN
self.linear1 = nn.Linear(256*62*62,64)

#RNN
self.lstm = torch.nn.LSTM(256, 10,bidirectional = True)

#output
self.linear2 = nn.Linear(64,num_of_chars)

def forward(self,x,input_size):
    
     
    x = self.conv1(input_size)
    x = nn.BatchNorm2d(x)
    x = F.relu(x)
    x = self.maxp(x)
    x = self.conv2(x)
    x = nn.BatchNorm2d(x)
    x = F.relu(x)
    x = self.maxp(x)
    x = self.dropout(x)
    x = self.conv3(x)
    x = nn.BatchNorm2d(x)
    x = F.relu(x)
    x = self.maxp(x)
    x = self.dropout(x)
    x = x.view((64,1024))
    x = self.linear1(x)
    
    x = self.lstm(x)
    x = self.lstm(x)
    
    x = self.linear2(x)
    x = nn.Softmax(x,dim=1)
    return x

---

但模型摘要完全不同。我被这些参数弄糊涂了。任何帮助都将不胜感激。如果你需要什么，告诉我。谢谢

你不明白的是什么？神经网络是一个操作链；通过某种操作，每一层都是前一层的图像。这转化为一个代码中的

inner=Activation（'relu'）（inner）

，另一个代码中的

x=F.relu（x）

。我很难理解输入数据是如何工作的，也就是说，如果它们的输入=输入（shape=（256,64,1），那么我应该执行x=x.view（256,64,1）或者别的什么。同样，在他们的例子中，lstm工作正常，在Pytorch中它要求隐藏的dim！。或多或少我被图层的形状和图像弄糊涂了。什么是形状弄糊涂了？你能举一个小得多的网络的例子吗？当前的一个有20个函数一个接一个地链接起来。我建议尝试使用3 fu开始是nctions，当你不那么困惑的时候，继续到20。你实际上没有问任何问题！你发布了一个相当长的代码和一个相当长的，应该是等效的代码，然后说你困惑了。好的，需要一些时间来弄清楚