Publication Date:
2023-11-06
Description:
Facial landmark detection on 3D scans is an active area of research. Most of the recent work focus on using statistical techniques and Deep Learning architectures to make the task of landmark detection an automated one. Recently, there has been an increasing interest in the field of Geometric Deep Learning that learns features directly from 3D data and offers a promising approach to improve the task of facial landmark detection. In this thesis, we investigate three different ways to improve an existing Geometric Deep Learning method for landmark detection on non-aligned face meshes. Our method offers promising results on meshes that are rotated up to 90° and outperforms the inter-rater variability obtained from the ground truth.
Language:
English
Type:
masterthesis
,
doc-type:masterThesis
