search

AaronJackson / vrn

:man: Code for "Large Pose 3D Face Reconstruction from a Single Image via Direct Volumetric CNN Regression"
http://aaronsplace.co.uk/papers/jackson2017recon/
MIT License
4.52k stars 742 forks source link

Performance? #34

Closed Marcophono2 closed 6 years ago

Marcophono2 commented 6 years ago

Hello!

A wonderful and creative project! I am only missing some informations about the performance. How long does it take to calculate a 3d model basing on a "standard" image depending on the gpu hardware? And: Is cpu also a strong factor? I am so impressed that I want to set up an own server to let your project run but would like to know a bit more about performance before deciding which components I buy.

Best regards Marc

AaronJackson commented 6 years ago

Hi Marc, thanks! On a Titan X the forward pass is very fast, only 300ms or so. Calculating the isosurface of the volume is a bit slower, maybe another 400-700ms depending on CPU, longer on slower machines. However you could speed it up by removing the last spatial upsampling layer and then applying the scaling to the 3D mesh (on XY only).

Marcophono2 commented 6 years ago

Thank you, Aaron! You are a sweetheart!

Best Marc