I have encountered some errors and need help

[wwweeeeiii]

Hello, I encountered some errors while replicating this project. When I use MERIT as my encoder, the following error occurs.

Later I find this problem in train() function, when using PVT decoder, the input image size will not change, while in MERIT model, the input image size will be changed to 256, while the original input image size is 256*0.75=196, I would like to ask how to solve this problem.