Help with model parameters

vision
1

data_transforms = {
‘train’: transforms.Compose([
transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
‘val’: transforms.Compose([
transforms.Resize(224),
transforms.CenterCrop(256),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
}

class Net(nn.Module):
def init(self):
super(Net, self).init()
self.conv1 = nn.Conv2d(3, 64, kernel_size=6, padding=2)
self.pool = nn.MaxPool2d(2, 2)
self.conv2 = nn.Conv2d(64, 128, kernel_size=6, padding=2)
self.fc1 = nn.Linear(256 * 6 * 6, 4096)
self.fc2 = nn.Linear(4096, 4096)
self.fc3 = nn.Linear(4096, 2)

def forward(self, x):
    x = self.pool(F.relu(self.conv1(x)))
    x = self.pool(F.relu(self.conv2(x)))
    x = x.reshape(x.size(0), -1)
    x = F.relu(self.fc1(x)) 
    x = F.relu(self.fc2(x))
    x = self.fc3(x)
    return x

RuntimeError: size mismatch, m1: [4 x 387200], m2: [9216 x 4096]

how can I set parameters for this model?

2

Since the number of features of the input activation to self.fc1 is 387200, you should set in_features of self.fc1 also to this value.
Your current model expects 256*6*6 input features, which doesn’t match the activation shape for your current input.

4

where is that 387200 is coming from

5

It’s coming from the number of features of the activation you are trying to pass to self.fc1.
I.e. x would have the shape here:

x = x.reshape(x.size(0), -1)
print(x.shape) # should be [batch_size, 387200]
x = F.relu(self.fc1(x))
6

So how would you write this model?

7

Either change in_features=387200 in self.fc1 as posted before.
Alternatively you could add more conv/pooling operation to reduce the spatial size of the activation or pass a smaller input tensor to the model.
The current shape is defined by your input shape and the conv/pool layers.