About the mismatch in the size of the training and inference inputs

[Albert22dai]

hello, I input the patch size of the network during training as (50,150,150), and use the model parameters reserved for training for inference, and the patch size of the input network during inference is (100,150,150), which will cause the inference output to be abnormal, does it have to strictly match the patch size of training and inference, which is unnecessary for CNN networks?
best wishes to you

[cabooster]

Hi Albert22dai. We always used the same patch size for training and inference. Only when the patch size is relatively small, different patch size will lead to different results. Since the receptive field of our network is 90, I think you can use a larger patch size.