samggggflynn

我在这里修正了。 I fixed the issue here. https://github.com/PeterL1n/BackgroundMattingV2/pull/200 https://github.com/samggggflynn/BackgroundMattingV2/blob/master/dataset/images.py

In my visualization of the results of changing the 'pose' parameters, the 6 parameters only include controls for the 'neck' and 'jaw'. Fisrt 3 dimensions for 'neck', last 3 dimensions...

> 使用绘世启动sd，不知道该怎么开启这个参数 --enable-insecure-extension-access，cmd_args.py里面有这个参数，但是不知道如何启用 just simply adding a character "--default=True" , after < "--enable-insecure-extension-access", action='store_true', >

> @mdv3101 any news on this? maybe you can try this https://github.com/harlanhong/ICCV2023-MCNET

> I met the problem too,I found there did not exsist any file or package had relation about 'lora_diffusion' @yjdqk try to install this repository -> https://github.com/cloneofsimo/lora

> I met the problem too,I found there did not exsist any file or package had relation about 'lora_diffusion' ``` git clone https://github.com/cloneofsimo/lora cd lora pip install --verbose -e ....

> Most likely it's because of your source video. Here is my result: > > qs_cn_half.mp4 nice job. 请问一下，你这个训练素材多少时长？ step1 和 step2 各训练了多少？

> 有什么tricks可以提高清晰度吗，感觉预测的图像清晰度相较训练集有所下降？ 遇到同样的问题，请问你找到原因或者解决了吗？增加数据和训练step有作用吗

may i ask more details, what does the face cropping region look like in your training data? Peter ***@***.***>于2024年5月13日 周一下午9:01写道： > Most likely it's because of your source video. Here...