search

Zz-ww / SadTalker-Video-Lip-Sync

本项目基于SadTalkers实现视频唇形合成的Wav2lip。通过以视频文件方式进行语音驱动生成唇形，设置面部区域可配置的增强方式进行合成唇形（人脸）区域画面增强，提高生成唇形的清晰度。使用DAIN 插帧的DL算法对生成视频进行补帧，补充帧间合成唇形的动作过渡，使合成的唇形更为流畅、真实以及自然。
1.79k stars 305 forks source link

加了DAIN补帧进行推理的话需要多大的显存？ | How much video memory do I need if I add DAIN frames for reasoning? #32

Open RuaYiii opened 1 year ago

RuaYiii commented 1 year ago

6g的显存完全可以胜任推理任务，但是加了DAIN后就炸显存了，换了10G的机子还是炸 想知道作者用的配置是多大显存进行推理的？

6g of video memory is fully capable of reasoning tasks, but after the addition of DAIN will blow up the memory, replaced by a 10G machine or fried I would like to know how much video memory the author used for the configuration for reasoning?

RobinBrackez commented 1 year ago

Have the same issue.

Out of memory error on GPU 0. Cannot allocate 558.000000MB memory on GPU 0, 7.999695GB memory has been allocated and available memory is only 0.000000B.

It's the only program running on gpu.

SuperMaximus1984 commented 1 year ago

Same here with 16 GB GPU (RTX3080), throws this error...

matthias33890 commented 8 months ago

Same here with 16 GB GPU

wangwenjie1314 commented 5 months ago
image image

同问，我用的云服务器 开启 use_DAIN 也是不行

yanhuo2020 commented 4 months ago

i have the same issue，i used 12G 3060，how can 1 resolve the issue