FacePoluke

我也遇到了，用cpu模式就好了，face_landmarker_v2_with_blendshapes这个任务只支持cpu跑

确保 utils/mp_models 下的模型没动过，然后mp_utils.py的这行 是CPU模式 https://github.com/Zejun-Yang/AniPortrait/blob/4e6c6aafc77d1285732008b536e025e5412a4550/src/utils/mp_utils.py#L24 这一行还有一个bug，把后面的[:600, 650:1350]删掉。https://github.com/Zejun-Yang/AniPortrait/blob/4e6c6aafc77d1285732008b536e025e5412a4550/scripts/preprocess_dataset.py#L99 其他地方我没有改动，可以正常跑通

你这看着好像没有报错，请检查一下你的图片格式，是jpg还是png。还有视频目录组织方式，这份代码适用input_dir下多个子目录，每个子目录下是一系列png图片。如果与这个格式不对，请自己修改代码适配

是的

The face landmark model of mediapipe is not robust for image with small face. You can try cropping the image to make the face larger.

> The face landmark model of mediapipe is not robust for image with small face. You can try cropping the image to make the face larger. You can also detect...

The video generated by audio-driven method is at 30 FPS. https://github.com/Zejun-Yang/AniPortrait/blob/e4601b8c023f0a9e46a4a6def2f524e9ded6cece/src/utils/audio_util.py#L20