A question about the CC dataset

[Aristot1e]

In CC dataset, some pixel values can be very large at about 4x10^4. This will result in, when I use the network structure within the paper to run under my own framework, the loss of values will be very large. But such a problem does not arise in the source code provided by the author. So I would like to ask the author whether the dataset has been preprocessed, and if it is convenient, can you tell me the method of processing?