Any step-by-step results available?

Thanks for sharing this great work :)

Just wonder if you have any step by step experiments for building the final loss formulation.

I tried to decompose the loss and see the effectiveness of each term, but it turns out that the loss works as a whole and sensitive to some part changes in my case.

Really appreciate your time.