Error to Change Image Resolution for Model Training

If the input image is 96, there will be no error. When I set it to 128, it should be an issue with the unt. I guess it was the network designed by the author according to 96 at that time? But the author also provided the image size parameter, how can it be solved? Thank you