what is the proper size of the input images?

[Coderxiaobai1]

Hi, big bro. i want to localize the abnormal area of my dataset, but the image in my dataset is 3000 pixels*4000 pixels, i wonder what is the proper solution of my problem. Should i slide the window to get patches to form the train and test set.
ps: The images are taken by uav from high to detect the transmission power pole is normal or not. I used to use object detection method, but it didn't perform well because there are so few abnormal poles. then i wonder if ganomaly method can apply to my problem. It haunts me for a long time. Can you spare some time to answer my question, please?😭😭😭

[tomatowithpotato]

same question, did you solve it?

[YRChen1998]

I tried different sizes. I found 128 is the max input size. You'd better set the isize in options.py ( isize<=128 and isize = 16*n).