Vehicle Matching Based on Similarity Metric Learning

With the development of new media technology,vehicle matching plays a further significant role in video surveillance systems.Recent methods explored the vehicle matching based on the feature extraction.Meanwhile,similarity metric learning also has achieved enormous progress in vehicle matching.But most of these methods are less effective in some realistic scenarios where vehicles usually be captured in different times.To address this cross-domain problem,we propose a cross-domain similarity metric learning method that utilizes theGANto generate vehicle imageswith another domain and propose the two-channel Siamese network to learn a similarity metric from both domains(i.e.,Day pattern or Night pattern)for vehicle matching.To exploit properties and relationships among vehicle datasets,we first apply the domain transformer to translate the domain of vehicle images,and then utilize the two-channel Siamese network to extract features from both domains for better feature similarity learning.Experimental results illustrate that our models achieve improvements over state-of-the-arts.

Learning an identity distinguishable space for large scale face recognition

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.

SSRE:Cell Type Detection Based on Sparse Subspace Representation and Similarity Enhancement

Accurate identification of cell types from single-cell RNA sequencing(scRNA-seq)data plays a critical role in a variety of scRNA-seq analysis studies.This task corresponds to solving an unsupervised clustering problem,in which the similarity measurement between cells affects the result significantly.Although many approaches for cell type identification have been proposed,the accuracy still needs to be improved.In this study,we proposed a novel single-cell clustering framework based on similarity learning,called SSRE.SSRE models the relationships between cells based on subspace assumption,and generates a sparse representation of the cell-to-cell similarity.The sparse representation retains the most similar neighbors for each cell.Besides,three classical pairwise similarities are incorporated with a gene selection and enhancement strategy to further improve the effectiveness of SSRE.Tested on ten real scRNA-seq datasets and five simulated datasets,SSRE achieved the superior performance in most cases compared to several state-of-the-art single-cell clustering methods.In addition,SSRE can be extended to visualization of scRNA-seq data and identification of differentially expressed genes.The matlab and python implementations of SSRE are available at https://github.com/CSUBioGroup/SSRE.