Inconsistent implementation with paper

[Luciennnnnnn]

Hi, I notice this codebase leverage p2 weighting strategy, and timestep sampling strategy is also different with the description in paper. Is my understanding right?

[BDeMo]

The branch: master was out-of-date and the p2 weighting strategy (could be the same issue as this one). Our implementation is mainly in branch: experiment, and the actual curves about sampling and weighting are implemented in speed.diffusion.speed.Speed_IDDPM. Both the timestep sampling strategy and p2 weighting strategy are the same as the description in our paper.

Now the master branch (default) is updated, merging branch: experiment, for reading convenience. If you are intereted in, Plz check it.