Purpose of reshaping the training labels

[kl-31]

I posted a similar question to the original unet repo. Here I'm referring to the following line in the create_train_data method in data.py:
train_label = np.reshape(train_label, (len(txt),self.out_cols * self.out_rows,self.num_class))

What is the purpose of this reshaping that converts the image out_cols, out_rows to a vector out_cols*out_rows? This does not seem to be necessary in the binary case. Is the multi-class case treated differently by the fit_generator method?

Thanks in advance for any insight.