CarrotZelda
CarrotZelda
我也遇到了这个问题，应该是未检测出来人脸的landmarks，我debug发现landmarks数组是空的。不知道怎么解决
我也遇到了这个问题，只需要更改一下docker的镜像就好了。 运行`sudo vi /etc/docker/daemon.json`，然后在daemon.json中添加以下内容后保存。 `{ "registry-mirrors": [ "https://docker.m.daocloud.io", "https://docker.nju.edu.cn", "https://dockerproxy.com" ] }` 最后重启docker：`sudo systemctl restart docker`。 此时就可以执行`docker compose up --build -d`了。 @tuoerning
我猜是那几个镜像链接失效了，我之前添加了好几个镜像没有冲突。你也可以试试把那几个去掉，加阿里云的镜像。
我也不清楚根本原因是什么，但是换镜像链接确实能够build成功，我昨晚在新环境上配了一下，添加了一些新的链接。build的过程很慢，但最终没有报错，build成功了。下面是我的daemon.json内的镜像内容。 "registry-mirrors": [ "https://docker.1panel.dev", "https://docker.fxxk.dedyn.io", "https://docker.xn--6oq72ry9d5zx.cn", "https://docker.m.daocloud.io", "https://a.ussh.net", "https://docker.zhai.cm", "https://docker.nju.edu.cn", "https://dockerproxy.com", "http://docker.1panel.live", "https://dockerpull.com", "https://docker.1panel.live", "https://dockerproxy.cn", "https://docker.hpcloud.cloud"],
还有一件事，我参考了issue89，https://github.com/YuqiHuai/SORA-SVL/issues/89#issuecomment-2370252260 把client文件夹与server文件夹中的Dockerfile做了些修改，在`RUN npm install`上面加了一行 `RUN npm config set registry https://registry.npmmirror.com`
I have obtained results similar to yours. So I checked the img_applied_mask during the training process and found that the mask almost covered the entire face.    
Could you please provide some guidance on how to limit the optimization to only the lower half of the face? Looking forward to your response. Thank you very much！ @AlonZolfi
Ok, thank you for your reply! Do you mean to lower the height of this new_mask_uv.png before training? 