DMD2 implementation

[Turlan]

Hi authors, thank you for open-sourcing this project. I was wondering if it would be possible to also release the DMD2 implementation. That would be very helpful for reproducing the comparisons presented in your paper. Thanks!

[zhengkw18]

Hi @Turlan, you can use the DMD implementation in this repo (which already includes the multi-step backward simulation trick in DMD2). The only missing part is an additional GAN objective, which we did not tune carefully, and did not observe a very significant advantage compared to not using GAN.