Lightweight Facial Landmark Detection Based on Multi-Branch Convolution and Re-Parameterization.
Saved in:
| Title: | Lightweight Facial Landmark Detection Based on Multi-Branch Convolution and Re-Parameterization. |
|---|---|
| Authors: | Liu, Qi1 laukei0305@163.com, Xu, Yang2 xuyang_1981@aliyun.com |
| Source: | Engineering Letters. Jul2026, Vol. 34 Issue 7, p2716-2727. 12p. |
| Subjects: | Parameterization, Convolutional neural networks, Machine learning, Computer vision, Artificial neural networks, Loss functions (Statistics) |
| Abstract: | Facial landmark detection is a core task in computer vision, widely applied in applications including facial recognition, emotion analysis, intelligent surveillance, and driver state monitoring. However, traditional methods often exhibit low accuracy under large pose variations, occlusions, and complex lighting conditions. To address these challenges, we propose a lightweight facial landmark detection algorithm, GMEW-PFLD, based on the improved PFLD model. First, we introduce the Ghost-MobileOne Bottleneck structure, which incorporates re-parameterization techniques to replace the conventional inverted residual block. This modification significantly reduces the model's parameter size and computational complexity while enhancing feature extraction efficiency and inference speed. Second, we embed a Efficient Multi-scale Attention (EMA) mechanism to effectively improve the model's ability to capture multi-scale facial details. Finally, we adopt the Wing Loss function, which adaptively scales gradients for small errors, mitigating gradient vanishing in this range and enhancing the model's robustness to outliers. Experimental results demonstrate that the improved model achieves reductions of 6.08%, 5.33%, and 5.42% in NME, ION, and IPN metrics, respectively, on the WFLW dataset. Simultaneously, the model size is reduced to only 12.5% of the original model, significantly lowering computational costs and reducing the failure rate by 28.2%. Overall, the experimental results show that the proposed algorithm achieves more accurate facial landmark detection with lightweight computational requirements, substantially improving the model's accuracy and robustness under challenging conditions. [ABSTRACT FROM AUTHOR] |
| Copyright of Engineering Letters is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
| Abstract: | Facial landmark detection is a core task in computer vision, widely applied in applications including facial recognition, emotion analysis, intelligent surveillance, and driver state monitoring. However, traditional methods often exhibit low accuracy under large pose variations, occlusions, and complex lighting conditions. To address these challenges, we propose a lightweight facial landmark detection algorithm, GMEW-PFLD, based on the improved PFLD model. First, we introduce the Ghost-MobileOne Bottleneck structure, which incorporates re-parameterization techniques to replace the conventional inverted residual block. This modification significantly reduces the model's parameter size and computational complexity while enhancing feature extraction efficiency and inference speed. Second, we embed a Efficient Multi-scale Attention (EMA) mechanism to effectively improve the model's ability to capture multi-scale facial details. Finally, we adopt the Wing Loss function, which adaptively scales gradients for small errors, mitigating gradient vanishing in this range and enhancing the model's robustness to outliers. Experimental results demonstrate that the improved model achieves reductions of 6.08%, 5.33%, and 5.42% in NME, ION, and IPN metrics, respectively, on the WFLW dataset. Simultaneously, the model size is reduced to only 12.5% of the original model, significantly lowering computational costs and reducing the failure rate by 28.2%. Overall, the experimental results show that the proposed algorithm achieves more accurate facial landmark detection with lightweight computational requirements, substantially improving the model's accuracy and robustness under challenging conditions. [ABSTRACT FROM AUTHOR] |
|---|---|
| ISSN: | 1816093X |
