How to use the pretrained model to test images with different sizes rather than 256 * 256 ?

[geonoon]

Hi,
Thanks for your repository.

 You provided the model  trained by using images of the size 256 * 256. 

  Can this pretrained model be used for predicting images with different sizes, e.g. 1024 * 1024, 712 * 543, or other sizes directly without pre-processing such as scaling or cropping?

  I tried the demo.py with image sizes rather than 256 * 256, but got failed without explicit error tips.

  Could you help in a quick reply?

   Thanks in advance