Can regression network training and testing of images with different sizes？

I trained a simple regression network to perform the image conversion task. In the training session, the size of the input image was set to a fixed value of 128*128. But in the test session, the input size is 1000*1000. The code is shown below：

inputSize = [128 128 1];
filterSize = 3;
numFilters = 128;


layers = [
    imageInputLayer(inputSize,'Normalization','rescale-zero-one','Min',-pi,'Max',pi)
    
    convolution2dLayer(filterSize,128,'DilationFactor',1,'Padding','same')
    batchNormalizationLayer
    reluLayer
    
    convolution2dLayer(filterSize,128,'DilationFactor',2,'Padding','same')
    batchNormalizationLayer
    reluLayer
    
    convolution2dLayer(filterSize,128,'DilationFactor',4,'Padding','same')
    batchNormalizationLayer
    reluLayer
    
    convolution2dLayer(1,1)
    regressionLayer]

options = trainingOptions('adam', ...
    'InitialLearnRate',0.001, ...
    'MaxEpochs',100, ...
    'MiniBatchSize',30, ...
    'Verbose',true,...
    'ValidationFrequency',30, ...
    'ValidationPatience',inf, ...
    'OutputNetwork','best-validation-loss', ...
    'ValidationData',{wrapVal,pVal});

[simplenet,info] = trainNetwork(wrapTrain,pTrain,layers,options);

An error message appears after the above code is run.

错误使用 DAGNetwork/predict

输入大小不正确。输入图像的大小必须为 [128 128 1]。

出错 SeriesNetwork/predict (第 302 行)

Y = this.UnderlyingDAGNetwork.predict(X, varargin{:});