Chapter 4 (Classifying images using deep CNNs)

[Abeltemi]

In the Training cell I am getting this error

RuntimeError: Calculated padded input size per channel: (1 x 1). Kernel size: (3 x 3). Kernel size can't be greater than actual input size

Training

model, loss_fn, optimizer = get_model()
accuracies, losses =[], []
epochs = 5
for epoch in range(epochs):
    epoch_losses, epoch_accuracies = [], []
    for ix, batch in enumerate(iter(trn_dl)):
        x, y = batch
        loss = train_batch(x, y, model, loss_fn, optimizer)
        epoch_losses.append(loss)
    epoch_loss = np.array(epoch_losses).mean()
    # accuracy check
    for ix, batch in enumerate(iter(trn_dl)):
        x, y = batch
        acc = accuracy_fn(x, y, model)
        epoch_accuracies.append(acc)
    epoch_acc = np.mean(epoch_accuracies)
    accuracies.append(epoch_acc)
    losses.append(epoch_loss)

CNN model architecture

# defining the CNN architecture
def get_model():
    model = nn.Sequential(
        nn.Conv2d(1, 64, kernel_size=3),
        nn.MaxPool2d(2),
        nn.ReLU(),
        nn.Conv2d(64, 128, kernel_size=3),
        nn.MaxPool2d(2),
        nn.ReLU(),
        nn.Flatten(),
        nn.Linear(3200, 256),
        nn.ReLU(),
        nn.Linear(256, 10)
    ).to(device)
    loss_fn = nn.CrossEntropyLoss()
    optimizer = Adam(model.parameters(), lr=1e-3)
    return model, loss_fn, optimizer

The Training and Accuracy functions

# creating a training function
def train_batch(x, y, model, loss_fn, opt):
    model.train()
    prediction = model(x)
    batch_loss = loss_fn(prediction, y)
    batch_loss.backward()
    opt.zero_grad()
    opt.step()
    return batch_loss.item()

@torch.no_grad()
def accuracy_fn(x, y, model):
    model.eval()
    prediction = model(x)
    max_val, argmaxes = prediction.max(-1)
    is_correct = argmaxes == y
    is_correct = is_correct.cpu().numpy().tolist()
    return is_correct

I have no idea what must have gone.. I Think the images was reduced to size smaller for the filter with 3x3. How do i fix this?

[mahmuudtolba]

1. Try to increase the size of the image to something like 224x224 through your dataset object
2. use padding with the CONV layer so that the shape of the image not decrease that much