Optimise inference time about video-retalking HOT 7 OPEN

Thanks understood, Step 6 always took most of inference time, any idea to speed it up, regards.

from video-retalking.

Thanks understood, Step 6 always took most of inference time, any idea to speed it up, regards.

Did you find any solution for Speed ?

from video-retalking.

报一下你现在硬件配置、视频尺寸和推理时间. 也许我可以给你一些建议。

from video-retalking.

Hi @zhangziliang04
I use 1 GPU Tesla T4 16GB, video Avatar size 832x960 (width, height).
To generate 8s video, it took 390 second in total.

Any idea to improve this, thanks

from video-retalking.

这个项目，默认有一个缓冲机制，第二次同一个驱动视频，会服用第一次训练的结果，具体对应1-4步骤，你可以看具体看一下。

from video-retalking.

@KartavyaBagga did you find yet? same question for @truongnhat-ryo @zhangziliang04 , how to make it faster ??

from video-retalking.

前面几步是有缓存，第二次用统一形象会有节约时间
但是，第六步占总时间的70～80%左右，是优化的关键
还需要多研究研究这步

from video-retalking.