Insufficient video memory，f1=0.0, loss=nan #38
Description
Hello, author, I encountered a problem: I used LEVIR-CD for training. Even though I adjusted the batchsize to a small size, the video memory was still not enough, so I cut the image to 256 and set batchsize=8. When training to the 119th epoch, the error of f1=0.0, loss=nan occurred. Please tell me how to solve the above problem.
There is another question. In the ph file, there is an image_size parameter setting, which corresponds to crop_size in utils. That is to say, input 1024 images and the images will be cropped before training. However, as mentioned before, directly inputting 1024 images will cause insufficient video memory, so I cropped the data set to 256 before inputting, but this caused the above problem.