How to increase the accuracy of model from chapter 04. PyTorch Custom Datasets

[nabeelalikhan0]

I tried many regularization techniques to fix this but...
Epoch: 50 | Train Loss: 0.5506 | Train Acc: 0.8003 | Test Loss: 1.6564 | Test Acc: 0.1875
Total training time: 44.556 seconds
that's the max i got

`
class Model_1(nn.Module):
def init(self,input_shape:int,hidden_units:int,output_shape:int):
super().init()
self.block_1 = nn.Sequential(
nn.Conv2d(input_shape,hidden_units,3,1,1),
nn.BatchNorm2d(hidden_units),
nn.ReLU(),
nn.Conv2d(hidden_units,hidden_units,3,1,1),
nn.BatchNorm2d(hidden_units),
nn.ReLU(),
nn.Conv2d(hidden_units,hidden_units,3,1,1),
nn.BatchNorm2d(hidden_units),
nn.ReLU(),
nn.MaxPool2d(2,2)
)

    self.block_2 = nn.Sequential(
        nn.Conv2d(hidden_units,hidden_units,3,1,1),
        nn.BatchNorm2d(hidden_units),
        nn.ReLU(),
        nn.Conv2d(hidden_units,hidden_units,3,1,1),
        nn.BatchNorm2d(hidden_units),
        nn.ReLU(),
        nn.Conv2d(hidden_units,hidden_units,3,1,1),
        nn.BatchNorm2d(hidden_units),
        nn.ReLU(),
        nn.MaxPool2d(2,2)
    )
    
    self.block_3 = nn.Sequential(
        nn.Conv2d(hidden_units,hidden_units,3,1,1),
        nn.BatchNorm2d(hidden_units),
        nn.ReLU(),
        nn.Conv2d(hidden_units,hidden_units,3,1,1),
        nn.BatchNorm2d(hidden_units),
        nn.ReLU(),
        nn.Conv2d(hidden_units,hidden_units,3,1,1),
        nn.BatchNorm2d(hidden_units),
        nn.ReLU(),
        nn.Dropout(0.25),
        nn.MaxPool2d(2,2)
    )

    self.classifier = nn.Sequential(
        nn.Flatten(),
        nn.LazyLinear(output_shape)
    )

def forward(self,x):
    x = self.block_1(x)
    x = self.block_2(x)
    x = self.block_3(x)
    x = self.classifier(x)
    return x

torch.manual_seed(42)
model = Model_1(3,10,len(classes_names)).to(device)
model

`

And this is my dataset
(Dataset ImageFolder Number of datapoints: 349 Root location: data\images\train StandardTransform Transform: Compose( Resize(size=(96, 96), interpolation=bilinear, max_size=None, antialias=True) TrivialAugmentWide(num_magnitude_bins=31, interpolation=InterpolationMode.NEAREST, fill=None) ToTensor() ), Dataset ImageFolder Number of datapoints: 100 Root location: data\images\test StandardTransform Transform: Compose( Resize(size=(96, 96), interpolation=bilinear, max_size=None, antialias=True) ToTensor() ))

Thankyou!

[mrdbourke]

Hi @nabeelalikhan0 ,

What does your dataset look like? Can you share some example images?

Have you tried increasing the size of the network? (e.g. adding more layers)

If that doesn't work you could try to use transfer learning and see if that helps: https://www.learnpytorch.io/06_pytorch_transfer_learning/