why train size is 96*96*96 ,but validate size in sliding_window_inference is 160*160*160?

[lg2578]

in train_transfoms,we can see the train patch is 96 ,96 ,96,but roi_size of sliding_window_inference is 160 ,160,160
why use different size here?

train_transforms = Compose(
    [
        LoadImaged(keys=["image", "label"]),
        EnsureChannelFirstd(keys=["image", "label"]),
        Spacingd(keys=["image", "label"], pixdim=(
            1.5, 1.5, 2.0), mode=("bilinear", "nearest")),
        Orientationd(keys=["image", "label"], axcodes="RAS"),
        ScaleIntensityRanged(
            keys=["image"], a_min=-57, a_max=164,
            b_min=0.0, b_max=1.0, clip=True,
        ),
        CropForegroundd(keys=["image", "label"], source_key="image"),
        RandCropByPosNegLabeld(
            keys=["image", "label"],
            label_key="label",
>>>    spatial_size=(96, 96, 96),
            pos=1,
            neg=1,
            num_samples=4,
            image_key="image",
            image_threshold=0,
        ),

code in validation step

if (epoch + 1) % val_interval == 0:
     model.eval()
     with torch.no_grad():
         for val_data in val_loader:
             val_inputs, val_labels = (
                 val_data["image"].to(device),
                 val_data["label"].to(device),
             )

>>>         roi_size = (160, 160, 160)
             sw_batch_size = 4
             val_outputs = sliding_window_inference(
                 val_inputs, roi_size, sw_batch_size, model)

[KumoLiu]

Hi @lg2578, since 96 and 160 are multiples of 32, there is no problem with the training process. And generally, the training size is smaller to improve the speed of training, when doing validation if you put a larger size may get a better model, but it is not the bigger the better.
Hope it can help you!
Thanks!