https://github.com/eeskimez/emotalkingface

报错

  1. 环境,pytorch和其他库之间版本对应的问题。
  2. cv2.getRotationMatrix2D TypeError: Can’t parse ‘center’. Sequence item with index 0 has a wrong type

解决:传入的参数必须是int类型,int32不行,套个int()

  1. librosa.load() File contains data in an unknown format.

在本地跑没问题,服务器上相同的环境报错。
最后发现是因为libroas读取文件时需执行ffmpeg这个命令,而服务器上没安装ffmpeg。使用sudo apt install ffmpeg指令安装。

  1. ffmpeg: error while loading shared libraries: libopenh264.so.5: cannot open shared object file: No such file or directory

在上一步安装的时候用过conda指令安装,要卸载。
Basically, I uninstalled conda’s version of ffmpeg(in the environment that’s having issues) using conda uninstall ffmpeg --force and (for Ubuntu), update the system wide version (I think this is what it’s called) with sudo apt install ffmpeg.

output

输入一张图片,自动截取头部并输出6段talking head video,每段各代表一种情绪。
输出被 resize 到128×128,分辨率很低。 ANG_generated.mp4_.mp4 (55.7KB)DIS_generated.mp4_.mp4 (54.95KB)FEA_generated.mp4_.mp4 (53.93KB)HAP_generated.mp4_.mp4 (56.39KB)NEU_generated.mp4_.mp4 (53.21KB)SAD_generated.mp4_.mp4 (54.32KB)