search

georgeretsi / smirk

Official Pytorch Implementation of SMIRK: 3D Facial Expressions through Analysis-by-Neural-Synthesis (CVPR 2024)
https://georgeretsi.github.io/smirk/
MIT License
181 stars 21 forks source link

Confidence #25

Open emlcpfx opened 2 months ago

emlcpfx commented 2 months ago

Is there anything within SMIRK that can tell you the confidence of how well the face is placed? Like a percentage of "Nailed It" to "Where is the face?!"

filby89 commented 1 week ago

Unfortunately we don't have something like this in terms of probability at least. One estimate would be to take the photometric/vgg loss of the generated image from the generator, although this is kinda a long shot and hard to quantify.