ZrrSkywalker
commented
5 months ago Thanks for your interest.
The original code can reproduce the result in our paper. You can refer to the provided several logs and ckpts based on our original code in the previous version of the repo. The new code can achieve much better results than those in the published paper as an enhanced variant.
The reproduction depends on many factors. We have tried our best to provide as much information as possible for reproduction. Thanks.
Hello, i find your more stable code have introduced new technique which are no longer consistent with your paper. Does this mean that the technology in the original paper cannot produce the published results? Additionally, we have tried the new code you have published, but still cannot produce the results you have published.