Making RNN for image classification

1

Hi guys,

I have array of rgb images with shape (3000, 3, 96, 96 ) with shape of labels like (3000, 4). labels are hot vector. but I am getting error like “Expected input batch_size (150) to match target batch_size (50)” when it tries to calculate the loss. would you please tell me what should i change?

network:

class RNNModel(nn.Module):
    def __init__(self, input_dim, hidden_dim, layer_dim, output_dim):
        super(RNNModel, self).__init__()
        # Hidden dimensions
        self.hidden_dim = hidden_dim

        # Number of hidden layers
        self.layer_dim = layer_dim

        # Building your RNN
        # batch_first=True causes input/output tensors to be of shape
        # (batch_dim, seq_dim, input_dim)
        # batch_dim = number of samples per batch
        self.rnn = nn.RNN(input_dim, hidden_dim, layer_dim, batch_first=True, nonlinearity='relu')

        # Readout layer
        self.fc = nn.Linear(hidden_dim, output_dim)

    def forward(self, x):
        # Initialize hidden state with zeros
        # (layer_dim, batch_size, hidden_dim)
        h0 = torch.zeros(self.layer_dim, x.size(0), self.hidden_dim).requires_grad_()

        # We need to detach the hidden state to prevent exploding/vanishing gradients
        # This is part of truncated backpropagation through time (BPTT)
        out, hn = self.rnn(x, h0.detach())

        out = self.fc(out[:, -1, :]) 
        return out

with parameters:

input_dim = 96
hidden_dim = 100
layer_dim = 1
output_dim = 4
model = RNNModel(input_dim, hidden_dim, layer_dim, output_dim)
criterion = nn.CrossEntropyLoss()
learning_rate = 0.01
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)  
print(len(list(model.parameters())))
print(list(model.parameters())[0].size())  # Input --> Hidden (A1)   torch.Size([100, 96])
print(list(model.parameters())[2].size())  # Input --> Hidden BIAS (B1)   torch.Size([100])
print(list(model.parameters())[1].size())  # Hidden --> Hidden (A3)   torch.Size([100, 100])
print(list(model.parameters())[3].size())  # Hidden --> Hidden BIAS(B3)  torch.Size([100])
print(list(model.parameters())[4].size())  # Hidden --> Output (A2)  torch.Size([4, 100])
print(list(model.parameters())[5].size())         # Hidden -> Output BIAS (B2)torch.Size([4])

with train loop of:

# Number of steps to unroll
seq_dim = 96  
num_epochs = 5
iter = 0
for epoch in range(num_epochs):
    for i, (images, labels) in enumerate(train_loader_fo):
        model.train()
        # Load images as tensors with gradient accumulation abilities
        images = images.float()
        images = images.view(-1, seq_dim, input_dim).requires_grad_()

        # Clear gradients w.r.t. parameters
        optimizer.zero_grad()

        # Forward pass to get output/logits
        # outputs.size() --> 100, 4
        outputs = model(images)

        # Calculate Loss: softmax --> cross entropy loss
        loss = criterion(outputs, labels)

        # Getting gradients w.r.t. parameters
        loss.backward()

        # Updating parameters
        optimizer.step()

        iter += 1

        if iter % 500 == 0:
            model.eval()
            # Calculate Accuracy         
            correct = 0
            total = 0
            # Iterate through test dataset
            for images, labels in vali_loader_fo:
                # Load images to a Torch tensors with gradient accumulation abilities
                images = images.view(-1, seq_dim, input_dim)

                # Forward pass only to get logits/output
                outputs = model(images)

                # Get predictions from the maximum value
                _, predicted = torch.max(outputs.data, 1)

                # Total number of labels
                total += labels.size(0)

                # Total correct predictions
                correct += (predicted == labels).sum()

            accuracy = 100 * correct / total

            # Print Loss
            print('Iteration: {}. Loss: {}. Accuracy: {}'.format(iter, loss.item(), accuracy))
2

nn.CrossEntropyLoss won’t work with one-hot encoded targets and expects a LongTensor in the shape [batch_size] containing the class indices.

However, since this is not the current error, could you please print the output and target shape before calculating the loss?

3

Hi, mi name is nicolas and i have a similar problem, in my case the rnn tensor is torch.Size([300, 7])
and labels torch.Size([100, 7])

4

Answered here.