Implementing face recognition efficiently to real world large scale dataset presents great challenges to existing approaches. The method in this paper was proposed to learn an identity distinguishable space for large ...Implementing face recognition efficiently to real world large scale dataset presents great challenges to existing approaches. The method in this paper was proposed to learn an identity distinguishable space for large scale face recognition in MSR-Bing image recognition challenge （IRC）. Firstly, a deep convolutional neural network （CNN） was used to optimize a 128 B embedding for large scale face retrieval. The embedding was trained via using triplets of aligned face patches from FaceScrub and CASIA-WebFace datasets. Secondly, the evaluation of MSR-Bing IRC was conducted according to a cross-domain retrieval scheme. The real-time retrieval in this paper was benefited from the K-means clustering performed on the feature space of training data. Furthermore, a large scale similarity learning （LSSL） was applied on the relevant face images for learning a better identity space. A novel method for selecting similar pairs was proposed for LSSL. Compared with many existing networks of face recognition, the proposed model was lightweight and the retrieval method was promising as well.展开更多
A discriminative local shape descriptor plays an important role in various applications.In this paper,we present a novel deep learning framework that derives discriminative local descriptors for deformable 3D shapes.W...A discriminative local shape descriptor plays an important role in various applications.In this paper,we present a novel deep learning framework that derives discriminative local descriptors for deformable 3D shapes.We use local"geometry images"to encode the multi-scale local features of a point,via an intrinsic parameterization method based on geodesic polar coordinates.This new parameterization provides robust geometry images even for badly-shaped triangular meshes.Then a triplet network with shared architecture and parameters is used to perform deep metric learning;its aim is to distinguish between similar and dissimilar pairs of points.Additionally,a newly designed triplet loss function is minimized for improved,accurate training of the triplet network.To solve the dense correspondence problem,an efficient sampling approach is utilized to achieve a good compromise between training performance and descriptor quality.During testing,given a geometry image of a point of interest,our network outputs a discriminative local descriptor for it.Extensive testing of non-rigid dense shape matching on a variety of benchmarks demonstrates the superiority of the proposed descriptors over the state-of-the-art alternatives.展开更多
基金supported by the Hi-Tech Research and Development Program of China ( 2013AA013303)the National Natural Science Foundation of China ( 61002011 )the Fundamental Research Funds for the Central Universities ( 2013RC1104)
文摘Implementing face recognition efficiently to real world large scale dataset presents great challenges to existing approaches. The method in this paper was proposed to learn an identity distinguishable space for large scale face recognition in MSR-Bing image recognition challenge （IRC）. Firstly, a deep convolutional neural network （CNN） was used to optimize a 128 B embedding for large scale face retrieval. The embedding was trained via using triplets of aligned face patches from FaceScrub and CASIA-WebFace datasets. Secondly, the evaluation of MSR-Bing IRC was conducted according to a cross-domain retrieval scheme. The real-time retrieval in this paper was benefited from the K-means clustering performed on the feature space of training data. Furthermore, a large scale similarity learning （LSSL） was applied on the relevant face images for learning a better identity space. A novel method for selecting similar pairs was proposed for LSSL. Compared with many existing networks of face recognition, the proposed model was lightweight and the retrieval method was promising as well.
基金partially funded by the National Key R&D Program of China(2018YFB2100602)the National Natural Science Foundation of China(61802406,61772523,61702488)+2 种基金Beijing Natural Science Foundation(L182059)the CCF–Tencent Open Research Fund,Shenzhen Basic Research Program(JCYJ20180507182222355)the Open Project Program of the State Key Lab of CAD&CG(A2004)Zhejiang University.
文摘A discriminative local shape descriptor plays an important role in various applications.In this paper,we present a novel deep learning framework that derives discriminative local descriptors for deformable 3D shapes.We use local"geometry images"to encode the multi-scale local features of a point,via an intrinsic parameterization method based on geodesic polar coordinates.This new parameterization provides robust geometry images even for badly-shaped triangular meshes.Then a triplet network with shared architecture and parameters is used to perform deep metric learning;its aim is to distinguish between similar and dissimilar pairs of points.Additionally,a newly designed triplet loss function is minimized for improved,accurate training of the triplet network.To solve the dense correspondence problem,an efficient sampling approach is utilized to achieve a good compromise between training performance and descriptor quality.During testing,given a geometry image of a point of interest,our network outputs a discriminative local descriptor for it.Extensive testing of non-rigid dense shape matching on a variety of benchmarks demonstrates the superiority of the proposed descriptors over the state-of-the-art alternatives.
