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
4.51k stars 746 forks source link

About training data #12

Closed matteosal closed 7 years ago

matteosal commented 7 years ago

First of all, congratulations for your very impressive results! A couple of questions: 1) Can you please clarify what's the training data for the vrn-unguided.t7 model released with this repo? In your paper, section 3.1, a method for generating data from the 300W-LP dataset is described. In section 4 you also present results on other datasets. 2) Do you plan to release such dataset at some point?

Thanks!

AaronJackson commented 7 years ago

Hey, thank you! The method was trained on images and voxelised meshes from 3DDFA-300W as in Section 3.1. No training was performed on AFLW2000, BU-4DFE or Florence, which are the three datasets we used for testing. Additionally, we did not train on the testing portions of Helen or LFPW (if I remember the names of the ones with test/train splits). I will speak to my supervisor about the possibility of releasing the pre-voxelised training set. Update: Afraid it's a no for now.