the model will converge faster than RAFT?

[Lyan-ing]

Since the model is initialized by the coarse flow instead of zero flow, will it converge faster than RAFT? Maybe it will get good performance with fewer training steps? thank you

[xiaofeng94]

Hey, we didn't evaluate the convergence rate, but you may reduce the update iterations by setting iters for this function. iters should work even you don't train with a specific iters.

[Lyan-ing]

Ok，thank u