Problem of training stages

[songzihang]

Hello ~ In the paper, the 1)G1, 2)G2 are trained separately and then 3)trained jointly. Finally 4)D1 is removed and G1,G2 are trained jointly with lower LR. There are four training stages. But in document of github, only three stages are proposed. I wonder how many stages truly. Can someone trained before help me ~ Thanks very much~