Why the default training epoch is as large as 500?

[LuffysMan]

The paper claimed only need 80 epochs to train. Another reproduction from seathiefwang has the same problem.

[LuffysMan]

I know. It is because of the stategy of data sampling. You sampled part of the whole dataset for one epoch.