Weakly Supervised Pretraining and Multi-Annotator Supervised Finetuning for Facial Wrinkle Detection

Analyzing extensive collections of images can be exceedingly resource-intensive if each facial wrinkle must be individually assessed. Moreover, the subjectivity inherent in manual segmentation processes can diminish the reliability of research findings and pose a substantial issue. To address this issue, we effectively combine wrinkle data labeled by multiple annotators to minimize inter-rater variability and utilize these image-label pairs for training our model.