Lumin

OK, we will work on it these months.

Sorry for the incorrect link. We use [DeepFashion2](https://github.com/switchablenorms/DeepFashion2) as stated in our paper, which contains 491K images. I think it is the reason why some images are missing.

In ``Usage``, we provide an example of the command that you mentioned. Please refer to it for more details. For training loss, it should decrease until convergence (maybe around 0.003)....

Thanks for your question. I am wondering whether all the annotations are inconsistent with the generated images or only some of them are missing. Would you mind providing some examples...