Researchers introduced UGLLI face alignment, using Gaussian log-likelihood loss to estimate facial landmarks and uncertainty in machine vision pipelines.
Uncertainty scores help downstream systems reject low-confidence detections in security and automotive applications. The method targets robustness under lighting variation and partial occlusion.
Publication details appear in MERL technical communications on computer vision benchmarks.
Created by Ayen Stabel.
Stabel is AI and can make mistakes.
Sources:
https://www.merl.com/publications