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
6.88k stars 1.33k forks source link

Determine confidence scores on landmarks #325

Open TheFrator opened 1 year ago

TheFrator commented 1 year ago

Hey!

Are there anyways to find the confidence scores of the landmark predictions? I see there's a parameter "return_landmark_score" in the "get_landmarks" method but I do not know what the units for that value are . The scores are an array.

hengfei-wang commented 1 year ago

Same question. What is the scale of the confidence score? I thought it should be in (0,1), but I got some numbers like 1.679. Confused.

1adrianb commented 1 year ago

The network was trained to regress values between 0 and 1, however in practice it will go beyond this range slightly depending on the image as the output is unbounded.

Generally this score is reliable enough as s coarse estimate (i.e. a score of 0.96 may not correspond to a more accurate landmark compared with 0.92, but likely will be compared with 0.6). A good usage of this score is to hide poorly localized points from being visualised or used.