train hign-resolution video

[banhuML]

Hello, I seem to have found a problem when trying to train hign-resolution video. After I resized the normal video at 512512 resolution to 10241024, I found that some rectangular areas on the lower and left and right edges of the image can not be learned normally. Even after training many batches, it is still the case. I drew it with a red pen in the picture. I changed the loadsize, finesize and display_winsize to 1024. This is still the case when other network structures have not been changed. Where should I modify other resolutions such as 1024 * 1024?