Inference with different input shape

[wincle]

Hi , thanks for your work, I have trained one model with image size 256*256, I could get good results in samples, but when I test it with other input shape, it get worse , do you know why it happened?

[taki0112]

How many shapes are there?

[wincle]

I change test_real_A from [1, 256,256 ,3] to [1,533,800,3] , which input image'size is 533*800.

[wincle]

https://www.dropbox.com/sh/vves0ghwvxzpmby/AAC9HqJL0cqBUcrtqLWchooya?dl=0 Here it is.

[qingmeizhujiu]

https://www.dropbox.com/sh/vves0ghwvxzpmby/AAC9HqJL0cqBUcrtqLWchooya?dl=0 Here it is.

Can not open?

[wincle]

I have fix it up, It's my mistake. Because I have changed the ratio of the trained images data.

[wujustin]

Hi , thanks for your work, I have trained one model with image size 256*256, I could get good results in samples, but when I test it with other input shape, it get worse , do you know why it happened?

@wincle The color of the result image is incorrect, can you share me some train set info or train detail info.

[szdree]

Can you talk about this random size implementation idea? Thank you。