General-to-specific learning for facial attribute classification in the wild

Sun, Yuechuan; Yu, Jun

doi:10.1016/j.jvcir.2018.09.003

Cited by 7 publications

(8 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In this subsection, on CelebA and LFWA datasets, we compare our method against MOON [6], Walk-and-Learn [36], SSP + SSG [9], General to Specific [7], DMTL [8] and MLGCN [13]. For making a fair comparison, we adopt the VGG16 [37] as our backbone, which is following the protocol in [7] and [6]. The performance comparison of the proposed solution with other state-of-the-arts on CelebA and LFWA is given in TABLE 7.…”

Section: Comparison With the State-of-the-art Methodsmentioning

confidence: 99%

“…The performance comparison of the proposed solution with other state-of-the-arts on CelebA and LFWA is given in TABLE 7. For [6], [7], [9], [36] and [8], we use the performance reported in their papers directly. For [13], the code was provided by the authors and we use it directly in our experiments.…”

Section: Comparison With the State-of-the-art Methodsmentioning

confidence: 99%

“…Han et al [8] tackle attribute correlation and heterogeneity with CNNs consisting of shared feature learning for all the attributes, and category-specific feature learning for heterogeneous attributes. Sun and Yu [7] propose a general-tospecific architecture which is able to implicitly discover the correlations of all the attributes considered while specifically focus on the distinctions. In this paper, the correlations are handled through GCN.…”

Section: B Classification-based Facial Attribute Recognitionmentioning

confidence: 99%

“…LFW contains 13,233 images of 5,749 people, [5] annotated these images with 40 attributes, which is named LFWA. We follow the protocol in [7] to split the dataset into 6283 training images and 6880 test images. In our experiments, we preprocess all images to the resolution of 120 × 150 which is followed by the alignment method introduced in [33].…”

Section: Experiments a Datasets And Settings 1) Datasetsmentioning

confidence: 99%

“…For instance, Liu et al [5] train independent classifiers for each attributes, which is able to generate discriminative attribute-specific feature representation. Rudy et al [6] utilize multivariate regression, Sun and Yu [7] and Han et al [8] employ general-to-specific methods, Kalayeh et al [9] and He et al [10] fuse additional segmentation or adversarial networks to model the relationships among attributes. However, they cannot address the three challenges mentioned above simultaneously.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Facial Attribute Recognition With Feature Decoupling and Graph Convolutional Networks

et al. 2019

View full text Add to dashboard Cite

The objective of facial attribute recognition is to predict a set of labels for each face image. Similar to general multi-label image classification task, facial attribute recognition suffers from several typical problems: discriminative feature learning, handling the correlations among attributes, and imbalanced training data. In this paper, we propose a unified facial attribute recognition solution via feature decoupling and graph convolutional networks (GCN): 1) We utilize the orthonormal regularizer to constrain that each dimension of the general facial representation represents a certain visual pattern. Meanwhile, a learnable matrix is unveiled, which can convert general facial feature into attribute-specific representation through modeling the importance of different visual patterns. 2) To handle the correlations between facial attributes, we build a GCN to capture the label dependencies and map the nodes in the proposed GCN to a set of interdependent attribute classifiers. Besides, we normalize the weights in all classifiers to alleviate the influence of data imbalance. We have conducted extensive experiments on two benchmarks, and both the qualitative and quantitative evaluation results have demonstrated the effectiveness of the proposed method.

show abstract