I am trying to reproduce the multi-input neural network of this tutorial: tutorial. This article uses PyTorch Lightning, while I want to use PyTorch, so I am adapting to my case. Basically I created my dataloaders and my network:
# Define loaders
from torch.utils.data import DataLoader
train_loader = DataLoader(train_set, batch_size=64, num_workers=2, drop_last=True, shuffle=True)
val_loader = DataLoader(val_set, batch_size=64, num_workers=2, drop_last=False, shuffle=False)
test_loader = DataLoader(test_set, batch_size=64, num_workers=2, drop_last=False, shuffle=False)
def conv_block(input_size, output_size):
block = nn.Sequential(
nn.Conv2d(input_size, output_size, (3, 3)), nn.BatchNorm2d(output_size), nn.ReLU(), nn.MaxPool2d((2, 2)),
)
return block
class SimpleCNN(nn.Module):
#Constructor
def __init__(self):
# Call parent contructor
super().__init__()
self.conv1 = conv_block(3, 16)
self.conv2 = conv_block(16, 32)
self.conv3 = conv_block(32, 64)
self.ln1 = nn.Linear(64 * 26 * 26, 16)
self.relu = nn.ReLU()
self.batchnorm = nn.BatchNorm1d(16)
self.dropout = nn.Dropout2d(0.5)
self.ln2 = nn.Linear(16, 5)
self.ln4 = nn.Linear(5, 10)
self.ln5 = nn.Linear(10, 10)
self.ln6 = nn.Linear(10, 5)
self.ln7 = nn.Linear(10, 1)
# Forward
def forward(self, img, tab):
img = self.conv1(img)
img = self.conv2(img)
img = self.conv3(img)
img = img.reshape(img.shape[0], -1)
img = self.ln1(img)
img = self.relu(img)
img = self.batchnorm(img)
img = self.dropout(img)
img = self.ln2(img)
img = self.relu(img)
tab = self.ln4(tab)
tab = self.relu(tab)
tab = self.ln5(tab)
tab = self.relu(tab)
tab = self.ln6(tab)
tab = self.relu(tab)
x = torch.cat((img, tab), dim=1)
x = self.relu(x)
return self.ln7(x)
Now, I try to pass a sample to my network to check if it works correctly, in order to start training later:
# Create the model
model = SimpleCNN()
img_x, tab_x, label_x = train_set[0]
print(img_x.shape, tab_x, label_x)
img_x
img_x = img_x.unsqueeze(dim=0)
output = model(img_x, tab_x)
output.shape
But here, I have this error: ValueError: Expected more than 1 value per channel when training, got input size torch.Size([1, 16]) at this line:
----> 4 output = model(img_x, tab_x)
........
---> 37 img = self.batchnorm(img)
I know that probably is because I am passing only one element to my network (I am not sure). However, If I remove that line from my neural network, so basically changing the forward method in this way:
def forward(self, img, tab):
img = self.conv1(img)
img = self.conv2(img)
img = self.conv3(img)
img = img.reshape(img.shape[0], -1)
img = self.ln1(img)
img = self.relu(img)
img = self.dropout(img)
img = self.ln2(img)
img = self.relu(img)
tab = self.ln4(tab)
tab = self.relu(tab)
tab = self.ln5(tab)
tab = self.relu(tab)
tab = self.ln6(tab)
tab = self.relu(tab)
x = torch.cat((img, tab), dim=1)
x = self.relu(x)
return self.ln7(x)
then I have this other error: RuntimeError: Tensors must have same number of dimensions: got 2 and 1 but this time referred to this line: ---> 49 x = torch.cat((img, tab), dim=1)
Edit: I solved the second error RuntimeError: Tensors must have same number of dimensions: got 2 and 1: basically I needed to unsqueeze also the tab_x in input:
model = SimpleCNN()
img_x, tab_x, label_x = train_set[0]
img_x = img_x.unsqueeze(dim=0)
tab_x = tab_x.unsqueeze(dim=0)
output = model(img_x, tab_x)
output.shape
and my output shape is torch.Size([1, 1]). However, I am still not able to fix the first error (the one in title)