The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
A Predição da Beleza Facial (FBP) é uma tarefa significativa de reconhecimento de padrões que visa alcançar uma avaliação consistente da atratividade facial com a percepção humana. Atualmente, Redes Neurais Convolucionais (CNNs) se tornaram o método principal para FBP. O objetivo de treinamento da maioria das CNNs convencionais é geralmente aprender núcleos de convolução estáticos, o que, no entanto, torna a rede bastante difícil de capturar informações de atenção global e, portanto, geralmente ignora as principais regiões faciais, por exemplo, olhos e nariz. Para resolver este problema, desenvolvemos uma nova maneira de convolução, Convolução Atenta Dinâmica (DyAttenConv), que integra o mecanismo de dinâmica e atenção na convolução em nível de kernel, com o objetivo de impor dinamicamente os kernels de convolução adaptados a cada face. DyAttenConv é um módulo plug-and-play que pode ser combinado de forma flexível com as arquiteturas CNN existentes, tornando a aquisição de recursos relacionados à beleza de forma mais global e cuidadosa. Extensos estudos de ablação mostram que nosso método é superior a outros mecanismos de fusão e atenção, e a comparação com outros estados da arte também demonstra a eficácia do DyAttenConv na tarefa de previsão da beleza facial.
Zhishu SUN
Fuzhou University
Zilong XIAO
Fuzhou University
Yuanlong YU
Fuzhou University
Luojun LIN
Fuzhou University
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copiar
Zhishu SUN, Zilong XIAO, Yuanlong YU, Luojun LIN, "Dynamic Attentive Convolution for Facial Beauty Prediction" in IEICE TRANSACTIONS on Information,
vol. E107-D, no. 2, pp. 239-243, February 2024, doi: 10.1587/transinf.2023EDL8058.
Abstract: Facial Beauty Prediction (FBP) is a significant pattern recognition task that aims to achieve consistent facial attractiveness assessment with human perception. Currently, Convolutional Neural Networks (CNNs) have become the mainstream method for FBP. The training objective of most conventional CNNs is usually to learn static convolution kernels, which, however, makes the network quite difficult to capture global attentive information, and thus usually ignores the key facial regions, e.g., eyes, and nose. To tackle this problem, we devise a new convolution manner, Dynamic Attentive Convolution (DyAttenConv), which integrates the dynamic and attention mechanism into convolution in kernel-level, with the aim of enforcing the convolution kernels adapted to each face dynamically. DyAttenConv is a plug-and-play module that can be flexibly combined with existing CNN architectures, making the acquisition of the beauty-related features more globally and attentively. Extensive ablation studies show that our method is superior to other fusion and attention mechanisms, and the comparison with other state-of-the-arts also demonstrates the effectiveness of DyAttenConv on facial beauty prediction task.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2023EDL8058/_p
Copiar
@ARTICLE{e107-d_2_239,
author={Zhishu SUN, Zilong XIAO, Yuanlong YU, Luojun LIN, },
journal={IEICE TRANSACTIONS on Information},
title={Dynamic Attentive Convolution for Facial Beauty Prediction},
year={2024},
volume={E107-D},
number={2},
pages={239-243},
abstract={Facial Beauty Prediction (FBP) is a significant pattern recognition task that aims to achieve consistent facial attractiveness assessment with human perception. Currently, Convolutional Neural Networks (CNNs) have become the mainstream method for FBP. The training objective of most conventional CNNs is usually to learn static convolution kernels, which, however, makes the network quite difficult to capture global attentive information, and thus usually ignores the key facial regions, e.g., eyes, and nose. To tackle this problem, we devise a new convolution manner, Dynamic Attentive Convolution (DyAttenConv), which integrates the dynamic and attention mechanism into convolution in kernel-level, with the aim of enforcing the convolution kernels adapted to each face dynamically. DyAttenConv is a plug-and-play module that can be flexibly combined with existing CNN architectures, making the acquisition of the beauty-related features more globally and attentively. Extensive ablation studies show that our method is superior to other fusion and attention mechanisms, and the comparison with other state-of-the-arts also demonstrates the effectiveness of DyAttenConv on facial beauty prediction task.},
keywords={},
doi={10.1587/transinf.2023EDL8058},
ISSN={1745-1361},
month={February},}
Copiar
TY - JOUR
TI - Dynamic Attentive Convolution for Facial Beauty Prediction
T2 - IEICE TRANSACTIONS on Information
SP - 239
EP - 243
AU - Zhishu SUN
AU - Zilong XIAO
AU - Yuanlong YU
AU - Luojun LIN
PY - 2024
DO - 10.1587/transinf.2023EDL8058
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E107-D
IS - 2
JA - IEICE TRANSACTIONS on Information
Y1 - February 2024
AB - Facial Beauty Prediction (FBP) is a significant pattern recognition task that aims to achieve consistent facial attractiveness assessment with human perception. Currently, Convolutional Neural Networks (CNNs) have become the mainstream method for FBP. The training objective of most conventional CNNs is usually to learn static convolution kernels, which, however, makes the network quite difficult to capture global attentive information, and thus usually ignores the key facial regions, e.g., eyes, and nose. To tackle this problem, we devise a new convolution manner, Dynamic Attentive Convolution (DyAttenConv), which integrates the dynamic and attention mechanism into convolution in kernel-level, with the aim of enforcing the convolution kernels adapted to each face dynamically. DyAttenConv is a plug-and-play module that can be flexibly combined with existing CNN architectures, making the acquisition of the beauty-related features more globally and attentively. Extensive ablation studies show that our method is superior to other fusion and attention mechanisms, and the comparison with other state-of-the-arts also demonstrates the effectiveness of DyAttenConv on facial beauty prediction task.
ER -