# Monocular 3D Face Reconstruction Using 3D Morphable Model and ElasticFace
> Taqwa A.S.A.

URL kanonis: https://discover.unhas.ac.id/publications/monocular-3d-face-reconstruction-using-3d-morphable-model-and-elasticface
Jurnal / Konferensi: Proceedings of the 2023 IEEE International Conference on Industry 4 0 Artificial Intelligence and Communications Technology Iaict 2023
Tahun terbit: 2023
DOI: https://doi.org/10.1109/IAICT59002.2023.10205588
Citations: 3

## Authors
- Taqwa A.S.A.

## Abstract
3D Morphable Model, one of the models used to reconstruct 3D face from 2D monocular image of face, has achieved satisfactory results along with computer vision and graphics development. However, reconstructing 3D face using a 3D Morphable Model in a weakly-supervised manner has its challenges because it does not require labels as ground truth and only relies on the similarity of features between 2D monocular image and 3D face. This research uses weakly-supervised 3D face reconstruction by comparing identity feature extraction. In this case, deep face recognition techniques used for identity feature extraction are ArcFace, CosFace, and ElasticFace. The 3D face reconstruction process is divided into 1) rigid fitting to fit the 3D face landmarks into face landmarks of 2D monocular image and 2) non-rigid fitting feature similarity with hybrid-level weak supervision applying diverse deep face recognition models. The results of the reconstruction are subsequently evaluated using the NoW challenge. Experimental results on the NoW protocol show that ElasticFace-Arc is the best deep face recognition for monocular 3d face reconstruction.

## Keywords
- Artificial intelligence
- Computer science
- Face (sociological concept)
- Computer vision
- Monocular
- Feature extraction
- Similarity (geometry)
- Feature (linguistics)
- Facial recognition system
- Pattern recognition (psychology)
- 3D reconstruction
- Iterative reconstruction
- Image (mathematics)
- Linguistics
- Social science
- Sociology
- Philosophy

---
Sumber: Discover Unhas — RIMS Universitas Hasanuddin.
Saat mengutip, gunakan DOI bila tersedia atau URL kanonis di atas.