search

1adrianb / face-alignment

:fire: 2D and 3D Face alignment library build using pytorch
https://www.adrianbulat.com
BSD 3-Clause "New" or "Revised" License
6.94k stars 1.33k forks source link

Recommended resolution for head crops? #176

Closed RSKothari closed 4 years ago

RSKothari commented 4 years ago

What resolution would you recommend I pass your module for head crops?

1adrianb commented 4 years ago

The face alignment network takes as input a 256x256px image, out of which the face is around 200-220px height. That detector doesn't care too much about resolution as long as its not very small or extremely large (ie >1000px). Of course, smaller images will be faster at least during the face detection stage.

RSKothari commented 4 years ago

Thanks. In what range does the detector return the face crops? Is it normalized to the Surrey face model?

1adrianb commented 4 years ago

No, there is no facial modeling involved. The detection themselves are following the Wider face dataset distribution while the crop adjustment its chosen empirically.