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

How to find landmark position in volume? #17

Closed hdjsjyl closed 6 years ago

hdjsjyl commented 6 years ago

For every volume, there is correspondence between landmark and vertices of volume?

AaronJackson commented 6 years ago

No, we discuss this limitation in the paper so please check there to more details.

On 26 October 2017 22:25:16 CEST, hdjsjyl notifications@github.com wrote:

For every volume, there is correspondence between landmark and vertices of volume?

-- Sent from my Android device with K-9 Mail. Please excuse my brevity.

hdjsjyl commented 6 years ago

@AaronJackson , In the paper, there is a sentence"The error is calculated on the face region only on approximately 19,000 vertices per facial mesh.", how to get this 19000 vertices in volume? what's the rule to select?

AaronJackson commented 6 years ago

You can find a MAT file containing a logical array defining which vertices we compute error on from the 3DDFA model. The original 3DDFA mesh will have 53215 vertices - if you have a different number, you are using the wrong groundtruth mesh (available in their code). When you plot the selected points, make sure they actually cover the front of the face and aren't scattered randomly around the head. www.cs.nott.ac.uk/~psxasj/download.php?file=vrn-3ddfa-vertexfilter