Mingjia Li

Well, I made a mistake in reading images (pngs) with PIL.Image instead of imageio while doing multiprocessing, sorry for the misreport. By the way, I compare the contents of provided...

Rereading the code, I found the error do exist in the provided datasets/generate_synthia_label.py within [line 75](https://github.com/JDAI-CV/FADA/blob/master/datasets/generate_synthia_label_info.py#L75), which should be like this one in [line 61](https://github.com/JDAI-CV/FADA/blob/master/datasets/generate_synthia_label_info.py#L61). Best regards.

To support my thoughts, I further modify the code in `ddim_loop` into: ```python class NullInversion: ... @torch.no_grad() def ddim_loop(self, latent): uncond_embeddings, cond_embeddings = self.context.chunk(2) all_latent = [latent] latent = latent.clone().detach()...

Thank you for the clear explanation! I am pleased to PR when I come up with a better comment.