Yumeng Li

Hi, thanks for your interest! This is similar to the implementation of Attend and Excite [here](https://github.com/huggingface/diffusers/blob/b69fd990ad8026f21893499ab396d969b62bb8cc/src/diffusers/pipelines/stable_diffusion_attend_and_excite/pipeline_stable_diffusion_attend_and_excite.py#L159). It just provides an option to disable saving attention layers for normal inference process...

"normal inference process" refers to standard Stable Diffusion process actually, then no attention map is needed for optimization. If you want to visualize the attention map, you would need to...

Hi @taluos , thanks for your interests. At the beginning of the project, we have tried on face datasets, which ISSA also results in better reconstruction quality. But the hyperparameters,...

Hi @taluos , 1. Yes, you could directly use a pretrained StyleGAN models. Just there wasn't one trained on Cityscapes, so I trained one myself. 2. Right, it contains images...

Hi @taluos 1. 50k images should be enough. 2. You might found this issue helpful :) https://github.com/boschresearch/ISSA/issues/7

Hi @jianlufu121 , thanks for your interest! Depending on the training dataset size, one can use a fixed training iteration. Some intermediate generation results are logged, you may check them...

Hi @samar-fathallah , thanks for your interest! Unfortunately, we are not able to release the segmentation related codes and models. But you could use [mmsegmentation](https://github.com/open-mmlab/mmsegmentation) to train and test semantic...