我是深度学习的新手，已经创建了一个模型来对我的图像进行分类。
目前，这个项目在Google Colab或Kaggle (CPU和GPU)上引发错误，但在我的个人电脑(带CPU)上没有。
模型:
class CNN(nn.Module):
def __init__(self):
super(CNN,self).__init__()
self.network1 = nn.Sequential(
nn.Conv2d(3, 32, kernel_size = 3, padding = 1),
nn.ReLU(),
nn.Conv2d(32, 64, kernel_size = 3, stride = 1, padding = 1),
nn.ReLU(),
nn.MaxPool2d(2,2),
# nn.AdaptiveAvgPool2d((128,128)),
nn.Conv2d(64, 128, kernel_size = 3, stride = 1, padding = 1),
nn.ReLU(),
nn.Conv2d(128 ,128, kernel_size = 3, stride = 1, padding = 1),
nn.ReLU(),
nn.MaxPool2d(2,2),
nn.Conv2d(128, 256, kernel_size = 3, stride = 1, padding = 1),
nn.ReLU(),
nn.Conv2d(256, 256, kernel_size = 3, stride = 1, padding = 1),
nn.ReLU(),
nn.AdaptiveMaxPool2d((4,4))
)
self.network2 = nn.Sequential(
nn.Flatten(),
nn.Linear(256*4*4, 1024),
nn.ReLU(),
nn.Linear(1024, 512),
nn.ReLU(),
nn.Linear(512, n_classes)
)
self.sigm = nn.Sigmoid()
def forward(self,x):
x = self.network1(x)
x = self.network2(x)
return self.sigm(x)
epoch = 0
model.train()
criterion = nn.BCELoss()
while True:
batch_losses = []
for imgs, labels in data:
imgs, labels = imgs.float().to(device), labels.to(device)
optimizer.zero_grad()
model_result = model(imgs)
loss = criterion(model_result, labels.type(torch.float))
batch_loss_value = loss.item()
loss.backward()
optimizer.step()
batch_losses.append(batch_loss_value)
loss_value = np.mean(batch_losses)
print("epoch:{:2d} iter:{:3d} train: loss:{:.3f}".format(epoch, iteration, loss_value))
if epoch % SAVE_FREQ == 0:
checkpoint_save(model, epoch)
epoch += 1
if EPOCHS < epoch:
break
错误:
RuntimeError Traceback (most recent call last)
/tmp/ipykernel_33/872363799.py in <module>
14 optimizer.zero_grad()
15
---> 16 model_result = model(imgs)
17 loss = criterion(model_result, labels.type(torch.float))
18
/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
1108 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
1109 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110 return forward_call(*input, **kwargs)
1111 # Do not call functions when jit is used
1112 full_backward_hooks, non_full_backward_hooks = [], []
/tmp/ipykernel_33/1050848783.py in forward(self, x)
32 self.sigm = nn.Sigmoid()
33 def forward(self,x):
---> 34 x = self.network(x)
35 return self.sigm(x)
/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
1108 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
1109 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110 return forward_call(*input, **kwargs)
1111 # Do not call functions when jit is used
1112 full_backward_hooks, non_full_backward_hooks = [], []
/opt/conda/lib/python3.7/site-packages/torch/nn/modules/container.py in forward(self, input)
139 def forward(self, input):
140 for module in self:
--> 141 input = module(input)
142 return input
143
/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)
1108 if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks
1109 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1110 return forward_call(*input, **kwargs)
1111 # Do not call functions when jit is used
1112 full_backward_hooks, non_full_backward_hooks = [], []
/opt/conda/lib/python3.7/site-packages/torch/nn/modules/linear.py in forward(self, input)
101
102 def forward(self, input: Tensor) -> Tensor:
--> 103 return F.linear(input, self.weight, self.bias)
104
105 def extra_repr(self) -> str:
RuntimeError: mat1 and mat2 shapes cannot be multiplied (256x16 and 4096x1024)
输入形状为(3,406,565)这个形状的模型的总结是:
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 32, 406, 565] 896
ReLU-2 [-1, 32, 406, 565] 0
Conv2d-3 [-1, 64, 406, 565] 18,496
ReLU-4 [-1, 64, 406, 565] 0
MaxPool2d-5 [-1, 64, 203, 282] 0
Conv2d-6 [-1, 128, 203, 282] 73,856
ReLU-7 [-1, 128, 203, 282] 0
Conv2d-8 [-1, 128, 203, 282] 147,584
ReLU-9 [-1, 128, 203, 282] 0
MaxPool2d-10 [-1, 128, 101, 141] 0
Conv2d-11 [-1, 256, 101, 141] 295,168
ReLU-12 [-1, 256, 101, 141] 0
Conv2d-13 [-1, 256, 101, 141] 590,080
ReLU-14 [-1, 256, 101, 141] 0
AdaptiveAvgPool2d-15 [-1, 256, 4, 4] 0
Flatten-16 [-1, 4096] 0
Linear-17 [-1, 1024] 4,195,328
ReLU-18 [-1, 1024] 0
Linear-19 [-1, 512] 524,800
ReLU-20 [-1, 512] 0
Linear-21 [-1, 18] 9,234
Sigmoid-22 [-1, 18] 0
================================================================
Total params: 5,855,442
Trainable params: 5,855,442
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 2.63
Forward/backward pass size (MB): 712.84
Params size (MB): 22.34
Estimated Total Size (MB): 737.80
----------------------------------------------------------------
解决方案在我的例子中，问题是我没有考虑图像形状的批处理大小，并且我的数据集只有很少的灰度图像和很少的alpha通道。
那么您提供的错误与您的代码不一致。在错误中有:
/tmp/ipykernel_33/1050848783.py in forward(self, x)
32 self.sigm = nn.Sigmoid()
33 def forward(self,x):
---> 34 x = self.network(x)
35 return self.sigm(x)
但是你提供的代码是:
def forward(self,x):
x = self.network1(x)
x = self.network2(x)
在我的例子中，问题是首先我没有考虑图像形状的批处理大小，其次我的数据集只有很少的灰度图像和很少的alpha通道。