Binary file to train network

1

Hello,

I am new to pytorch and I am trying to import binary files (matlab matrices which are images) to train a standard resnet18 with. What is the best way to do this? Is there a way to import binary files with torchvision.datasets or do I have to write my own dataloader? Is it possible to train the network with these binary files or do I have to imoprt every file and transform it to a tensor on my own?

Thanks in advance.

2

You could use scipy.io.loadmat to load the MATLAB files.
Once you have loaded them as numpy arrays, you would have to transform them to tensors via torch.from_numpy().
I would recommend to write a custom Dataset as described here for this use case.