How to make dataset and dataloader from wavelet transformed image(npy.file)?

vision
1

Hello teachers.
I’m a starter for PyTorch and i’m thinking of building up CNN model for fault detection of vibration signals.

Before making pytorch dataset and dataloader, i created wavelet transformed image from signal.

1
1849×710 45.6 KB

  • type : numpy array
  • dimension : (100,50,50), 100 images with 50x50
  • value : around -0.01~0.01

after loading npy.file with np.load function,
I don’t know how to make my Dataset Class for Dataloader,
because it is not regular image.

Please help me :frowning:

Have a good day!

2

You don’t have to use regular images in your Dataset so your custom numpy arrays should be fine.
Here is a small dummy example:

class MyDataset(Dataset):
    def __init__(self, data):
        self.data = torch.from_numpy(data).float()
        
    def __getitem__(self, index):
        x = self.data[index]
        return x
    
    def __len__(self):
        return len(self.data)

I’m not sure, what your targets are, so you might want to pass them additionally to the Dataset or calculate them in __getitem__.

1 Like
3

Hi Ptrblck,
I really appreciate with your answer but there is still a problem.

3
31015×749 16.4 KB

Error says : object of type ‘type’ has no len()

  • The thing that i wanna do with this is, inputting images with 50x50 size and
    building Convolutional AutoEncoder model with unsupervised learning for anomaly detection.
  • I’m wondering i need to resize (100,30,30) into (100,1,30,30)… because input dimension is comprised of (batchsize, channel, height, width) in every the PyTorch Tutorials.

Thanks again!!

4

You want to pass an instance of the dataset to the dataloader constructor, not the type.

Best regards

Thomas

5

Thank you very much, tom,

but im afraid that i still don’t know how to modify the code for passing an instance for this case…

Regards,

David

6

i solved with dataset=MyDataset()

Thank you very much :slight_smile: