Is there a way to use pretrained networks on CIFAR10 without altering the Image size from 32 x 32 to 224 x 244?

vision
1

I would like to exploit the pre-trained networks on CIFAR10 but don’t want to alter the image size from 32 x 32 to 224 x 244.

2

For some models the input might be too small since some intermediate activations might be smaller than a specific conv kernel.
If you encounter these issues, you would have to increase the input up to a specific size until at least a single pixel is produced by the feature extractor.
Depending on the model architecture the minimal size might be smaller than 224x224.