I have a question!

[OSHMOS]

@Morpheus3000
I deeply respect your work.

Unlike previous IID models, this time I can infer without resizing to 256x256 image size. Is this because it is not trained with a single image size? To put it simply, can custom data be inferred as it is without resizing?

[Morpheus3000]

Hello,

Thanks for your interest in my work! Regarding the size issue, it is because the network is fully convolutional, so it is not dependent on the input size. As long as the bottleneck (the downscaling) works out to be more than 1 pixel, different resolution should not be a problem.