Jianzhu Guo

What about the running (inference) speed? @ZHANG-SHI-CHANG

这个 jekyll 主题确实挺漂亮，**但若打开主页后，迅速滑到底部，过1-2秒会有个明显的卡顿。** 主页地址：https://guojianzhu.com/ 这个可能是什么原因呢 🤦‍♀️？

I also found that RGB -> HSV conversion differs in PIL and OpenCV, the 1->1 conversion maybe impossible without modifying/re-writing OpenCV source code. I wrote a testing script for hue...

The same issue on [v4.16.1](https://github.com/coder/code-server/releases/tag/v4.16.1).

> I have uploaded `MeGlass_ori.zip` to [Kaggle](https://www.kaggle.com/datasets/mantasu/meglass) so now it's much easier to download for non-Baidu users. If it's okay with @cleardusk, I can keep it there with the same...

感谢反馈💗 @yingDev 问题可否描述的更清晰一些嘛？比如这个gif是什么参数组合跑出来的。 Thank you for your feedback! 💗 Could you describe the issue more clearly? For example, what parameter combination was used to generate this GIF?

@yingDev 可以尝试： ```bash python inference.py \ -s tmp/issue471.png \ -d tmp/issue471.png \ -o out \ --no_flag_do_crop \ --no_flag_relative_motion \ --no_flag_pasteback \ --no_flag_stitching \ --animation_region="all" ``` `tmp/issue471.png` 即上面的输入图。 输入vs.输出对比（GIF）： 

色差可以理解为网络有个“重建”+“超分”导致。 > @cleardusk 感谢回复。确实有效，也就是说禁止裁剪缝合。 但是色差可能是什么原因呢？ 我尝试修改源码输出为 png，还是明显看出颜色变暗了一点点。

Thanks for your report. @sixone-Jiang @huangxin168 Could you please provide an example, e.g., provide side-by-side comparisons ?

> 很抱歉我不能提供一个视频，但我可以给你提供我实验时的核心设置， 'flag_normalize_lip': False, 'flag_relative_motion': False, 'flag_do_crop': True, 'flag_pasteback': True, 'flag_stitching': True, 'animation_region': "lip", 'driving_option': "expression-friendly", 对于driving vedio，可以先对人脸下采样两倍，采取kernel=5的高斯模糊，再上采样到原分辨率用以获取低分辨率 It seems that the motion extractor is sensitive to the disturb of...