Advanced Industrial SCIence (산업과 과학)

Convergence Society for SMB (중소기업융합학회)
권호 표지
DOI QR코드

DOI QR Code

Enhanced Facial Landmark Detection (EFLD) with Lightweight and Efficient Networks

경량 및 효율적인 네트워크를 활용한 향상된 얼굴 랜드마크 검출 연구

  • Sunghyuck Hong (Division of Advanced IT, IoT major, Baekseok University)
  • 홍성혁 (백석대학교 첨단IT학부, IoT 전공)
  • Received : 2024.08.11
  • Accepted : 2025.01.20
  • Published : 2025.01.30
Download PDF
⟨ Previous Next ⟩

Abstract

This study introduces the Efficient Facial Landmark Detection (EFLD) model, which combines state-of-the-art network design and innovative loss methods to respond to various challenges arising from real facial images. EFLD presents methods to solve the problem of imbalance in learning data as well as global and local variations within faces. EFLD outperforms existing methods in terms of efficiency, accuracy, and compactness, achieving state-of-the-art results on standard benchmarks while requiring significantly fewer resources in terms of model size and processing speed. Extensive testing on datasets such as 300W and AFLW shows that EFLD consistently outperforms previous approaches in terms of accuracy and speed. Additionally, this study sets a new standard in the field of facial landmark detection by introducing a practical system optimized for mobile devices and suggests future development possibilities.

이 연구는 효율적인 얼굴 랜드마크 검출(Efficient Facial Landmark Detection, EFLD) 모델을 도입하며, 이는 실제 얼굴 이미지에서 발생하는 다양한 도전에 대응하기 위해 최첨단 네트워크 설계와 혁신적인 손실 방법을 결합한 모델이다. EFLD는 얼굴 내의 전역 및 국부적인 변동뿐만 아니라 학습 데이터의 불균형 문제를 해결하기 위한 방법들을 제시한다. EFLD는 효율성, 정확성 및 컴팩트성 면에서 현존하는 방법들을 능가하며, 표준 벤치마크에서 최첨단 결과를 달성하면서도 모델 크기와 처리 속도 면에서 현저히 적은 자원을 필요로 한다. 300W 및 AFLW와 같은 데이터셋에서의 광범위한 테스트 결과, EFLD는 이전 접근법에 비해 정확성과 속도 면에서 일관되게 우수함을 보인다. 또한, 이 연구는 모바일 기기에 최적화된 실용적인 시스템을 도입하여 얼굴 랜드마크 검출 분야에서 새로운 기준을 세우고 미래의 발전 가능성을 제시한다.

Keywords

Acknowledgement

This research was supported by 2024 Baekseok University research fund.

References

  1. Chandran, P., Zoss, G., Gotardo, P., & Bradley, D. (2023). Continuous landmark detection with 3D queries. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). https://studios.disneyresearch.com
  2. Guo, X., Li, S., Yu, J., Zhang, J., Ma, J., Ma, L., Liu, W., & Ling, H. (2019). PFLD: A practical facial landmark detector. arXiv preprint arXiv:1902.10859. https://arxiv.org/abs/1902.10859 10859
  3. Guo, X., Li, S., Yu, J., Zhang, J., Ma, J., Ma, L., Liu, W., & Ling, H. (2019). PFLD: A practical facial landmark detector. Papers With Code. https://paperswithcode.com/paper/pfld-a-practical-facial-landmark-detector
  4. Kar, P., Chudasama, V. M., Onoe, N., Wasnik, P., & Balasubramanian, V. (2023). Fiducial focus augmentation for facial landmark detection. In Proceedings of the 34th British Machine Vision Conference(BMVC). https://proceedings.bmvc2023.org
  5. Proll, S. (2023). Facial landmark detection is still easy with MediaPipe (2023 update). Retrieved from https://www.samproell.io
  6. Zeng, L., Chen, L., Bao, W., Li, Z., Xu, Y., Yuan, J., & Kalantari, N. K. (2023). 3D-aware facial landmark detection via multi-view consistent training on synthetic data. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR), 12747-12758. https://openaccess.thecvf.com/content/CVPR2023/papers/Zeng_3D-Aware_Facial_Landmark_Detection_via_Multi-View_Consistent_Training_on_Synthetic_CVPR_2023_paper.pdf
  7. Zhou, Z., Li, H., Liu, H., Wang, N., Yu, G., & Ji, R. (2023). STAR loss: Reducing semantic ambiguity in facial landmark detection. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). https://github.com/ZhenglinZhou/STAR
  8. Chandran, P., Zoss, G., Gotardo, P., & Bradley, D. (2023). Continuous landmark detection with 3D queries. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). https://cvpr.thecvf.com
  9. Guo, X., Li, S., Yu, J., Zhang, J., Ma, J., Ma, L., Liu, W., & Ling, H. (2019). PFLD: A practical facial landmark detector. arXiv Vanity. https://ar5iv.labs.arxiv.org/html/1902.10859
  10. Zeng, L., Chen, L., Bao, W., Li, Z., Xu, Y., Yuan, J., & Kalantari, N. K. (2023). 3D-aware facial landmark detection via multi-view consistent training on synthetic data. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). https://people.engr.tamu.edu
  11. Anonymous. (2023). Lite-HRNet Plus: Fast and accurate facial landmark detection. arXiv preprint arXiv:2308.12133. https://ar5iv.labs.arxiv.org/html/2308.12133
  12. Anonymous. (2023). STAR loss: Reducing semantic ambiguity in facial landmark detection. arXiv preprint arXiv:2306.02763. https://ar5iv.labs.arxiv.org/html/2306.02763
  13. Anonymous. (2023). Precise facial landmark detection by reference heatmap transformer. arXiv preprint arXiv:2303.07840. https://ar5iv.labs.arxiv.org/html/2303.07840
  14. Anonymous. (2023). KeyPosS: Plug-and-play facial landmark detection through GPS-inspired true-range multilateration. arXiv preprint arXiv:2305.16437. https://ar5iv.labs.arxiv.org/html/2305.16437
  15. Anonymous. (2023). 3D-aware facial landmark detection via multi-view consistent training on synthetic data. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR). https://openaccess.thecvf.com
  16. King, D. (n.d.). dlib: A toolkit for making real-world machine learning and data analysis applications in C++ [Computer software]. GitHub. https://github.com/davisking/dlib