search

YuanxunLu / LiveSpeechPortraits

Live Speech Portraits: Real-Time Photorealistic Talking-Head Animation (SIGGRAPH Asia 2021)
MIT License
1.15k stars 198 forks source link

How to run demo in "Real-time" #81

Open anthnyprschka opened 1 year ago

anthnyprschka commented 1 year ago

Hello, can someone guide me how to run demo in real-time, these are the latencies when I run in Colab notebook with GPU runtime:

Processing audio: 00083 ...
1. Computing APC features...
2. Manifold projection...
LLE projection: 100% 1374/1374 [**00:00**<00:00, 5131.87it/s]
3. Audio2Mouth inference...
4. Headpose inference...
generating headpose: 100% 672/672 [**00:06**<00:00, 111.27it/s]
5. Post-processing...
100% 672/672 [00:00<00:00, 17413.64it/s]
6. Image2Image translation & Saving results...
Image2Image translation inference: 100% 672/672 [**00:53**<00:00, 12.59it/s]

It takes around 0:59min in total...

RaiAmanRai commented 10 months ago

hey @anthnyprschka, would you be kind enough to share the colab notebook that you are using. I am trying to work out the latency, and i can try running it on A100 to see if that improves the latency!!

Thanks