about fine-tuning the loss function

[doomysday]

In the paper, the author say "we first train the network with MAE loss and then fine-tune it by MSE loss.",but I cannot find the clue about how to fine-tune the Loss function in the training phase, anybody know where it is in the code?