Making training dataset by other images

cleardusk / 3DDFA

The PyTorch improved version of TPAMI 2017 paper: Face Alignment in Full Pose Range: A 3D Total Solution.

MIT License

3.6k stars 651 forks source link

Making training dataset by other images #178

Open GUNGang opened 3 years ago

GUNGang commented 3 years ago

Hello,cleardusk,thanks for your sharing. Now I need to make a new dataset using my own images,but I don't know how to do it. The problems are: 1.How to get the 3DMM params from raw images, just like 300W-3D with 300W, by hand or code? 2.How to transfer 300W-3D to the training dataset in 3DDFA? I'm a newcomer in this filed, sorry for any possible trouble caused. Thanks again!

cleardusk commented 3 years ago

The 3DMM parameters are generated by analysis-by-synthesis. Parameters clipping and z-score normalization. You can refer to 3DDFA_V2 paper for more details.

Kakoedlinnoeslovo commented 3 years ago

Hi, thank you for amazing repo! Can you please provide more info about how to do custom param_all_norm.npy for my images? Thank you for reply!