自监督深度残差函数映射网络的3维模型对应关系计算

目的 针对传统非刚性3维模型的对应关系计算方法需要模型间真实对应关系监督的缺点,提出一种自监督深度残差函数映射网络（self-supervised deep residual functional maps network,SSDRFMN）。方法 首先将局部坐标系与直方图结合以计算3维模型的特征描述符,即方向直方图签名（signature of histograms of orientations,SHOT）描述符;其次将源模型与目标模型的SHOT描述符输入SSDRFMN,利用深度函数映射（deep functional maps,DFM）层计算两个模型间的函数映射矩阵,并通过模糊对应层将函数映射关系转换为点到点的对应关系;最后利用自监督损失函数计算模型间的测地距离误差,对计算出的对应关系进行评估。结果 实验结果表明,在MPI-FAUST数据集上,本文算法相比于有监督的深度函数映射（supervised deep functional maps,SDFM）算法,人体模型对应关系的测地误差减小了1.45;相比于频谱上采样（spectral upsampling,SU）算法减小了1.67。在TOSCA数据集上,本文算法相比于SDFM算法,狗、猫和狼等模型的对应关系的测地误差分别减小了3.13、0.98和1.89;相比于SU算法分别减小了2.81、2.22和1.11,并有效克服了已有深度函数映射方法需要模型间的真实对应关系来监督的缺点,使得该方法可以适用于不同的数据集,可扩展性大幅增强。结论 本文通过自监督深度残差函数映射网络训练模型的方向直方图签名描述符,提升了模型对应关系的准确率。本文方法可以适应于不同的数据集,相比传统方法,普适性较好。     

基于深度感知特征提取的室内场景理解

从深度图RGB-D域中联合学习RGB图像特征与3D几何信息有利于室内场景语义分割,然而传统分割方法通常需要精确的深度图作为输入,严重限制了其应用范围。提出一种新的室内场景理解网络框架,建立基于语义特征与深度特征提取网络的联合学习网络模型提取深度感知特征,通过几何信息指导的深度特征传输模块与金字塔特征融合模块将学习到的深度特征、多尺度空间信息与语义特征相结合,生成具有更强表达能力的特征表示,实现更准确的室内场景语义分割。实验结果表明,联合学习网络模型在NYU-Dv2与SUN RGBD数据集上分别取得了69.5%与68.4%的平均分割准确度,相比传统分割方法具有更好的室内场景语义分割性能及更强的适用性。     

Multiflash stereopsis: depth-edge-preserving stereo with small baseline illumination

Traditional stereo matching algorithms are limited in their ability to produce accurate results near depth discontinuities, due to partial occlusions and violation of smoothness constraints. In this paper, we use small baseline multi-flash illumination to produce a rich set of feature maps that enable acquisition of discontinuity preserving point correspondences. First, from a single multi-flash camera, we formulate a qualitative depth map using a gradient domain method that encodes object relative distances. Then, in a multiview setup, we exploit shadows created by light sources to compute an occlusion map. Finally, we demonstrate the usefulness of these feature maps by incorporating them into two different dense stereo correspondence algorithms, the first based on local search and the second based on belief propagation. Experimental results show that our enhanced stereo algorithms are able to extract high quality, discontinuity preserving correspondence maps from scenes that are extremely challenging for conventional stereo methods. We also demonstrate that small baseline illumination can be useful to handle specular reflections in stereo imagery. Different from most existing active illumination techniques, our method is simple, inexpensive, compact, and requires no calibration of light sources.     

部件检测和语义网络的细粒度鞋类图像检索

目的 细粒度图像检索是当前细粒度图像分析和视觉领域的热点问题。以鞋类图像为例,传统方法仅提取其粗粒度特征且缺少关键的语义属性,难以区分部件间的细微差异,不能有效用于细粒度检索。针对鞋类图像检索大多基于简单款式导致检索效率不高的问题,提出一种结合部件检测和语义网络的细粒度鞋类图像检索方法。方法 结合标注后的鞋类图像训练集对输入的待检鞋类图像进行部件检测;基于部件检测后的鞋类图像和定义的语义属性训练语义网络,以提取待检图像和训练图像的特征向量,并采用主成分分析进行降维;通过对鞋类图像训练集中每个候选图像与待检图像间的特征向量进行度量学习,按其匹配度高低顺序输出检索结果。结果 实验在UT-Zap50K数据集上与目前检索效果较好的4种方法进行比较,检索精度提高近6%。同时,与同任务的SHOE-CNN（semantic hierarchy of attribute convolutional neural network）检索方法比较,本文具有更高的检索准确率。结论 针对传统图像特征缺少细微的视觉描述导致鞋类图像检索准确率低的问题,提出一种细粒度鞋类图像检索方法,既提高了鞋类图像检索的精度和准确率,又能较好地满足实际应用需求。     

Correspondence consensus of two sets of correspondences through optimisation functions

We present a consensus method which, given the two correspondences between sets of elements generated by separate entities, enounces a final correspondence consensus considering the existence of outliers. Our method is based on an optimisation technique that minimises the cost of the correspondence while forcing (to the most) to be the mean correspondence of the two original correspondences. The method decides the mapping of the elements that the original correspondences disagree and returns the same element mapping when both correspondences agree. We first show the validity of the method through an experiment in ideal conditions based on palmprint identification, and subsequently present two practical experiments based on image retrieval.     

Tracking multiple features using relaxation

A new algorithm is introduced for tracking multiple features in an image sequence. First, the proposed method iteratively reduces the disparity of each possible match by relaxation labeling. It is assumed that all trajectories are smooth and the smoothness is used as the measure for correspondence. Some cases of wrong correspondences can be recovered by a proposed scheme called constraint-aided exchange during the tracking process. Occluded or missing feature points can be detected and predicted in the proposed algorithm. Finally, the algorithm is applied to data obtained from real world scenes. The human motion analysis can be achieved by the tracking algorithm.     

结合改进密集模块深度估计网络和多视几何的视觉里程计

以多视图几何原理为基础,有效结合卷积神经网络进行图像深度估计和匹配筛选,构造无监督单目视觉里程计方法.针对主流深度估计网络易丢失图像浅层特征的问题,构造一种基于改进密集模块的深度估计网络,有效地聚合浅层特征,提升图像深度估计精度.里程计利用深度估计网络精确预测单目图像深度,利用光流网络获得双向光流,通过前后光流一致性原则筛选高质量匹配.利用多视图几何原理和优化方式求解获得初始位姿和计算深度,并通过特定的尺度对齐原则得到全局尺度一致的6自由度位姿.同时,为了提高网络对场景细节和弱纹理区域的学习能力,将基于特征图合成的特征度量损失结合到网络损失函数中.在KITTI Odometry数据集上进行实验验证,不同阈值下的深度估计取得了85.9%、95.8%、97.2%的准确率.在09和10序列上进行里程计评估,绝对轨迹误差在0.007 m.实验结果验证了所提出方法的有效性和准确性,表明其在深度估计和视觉里程计任务上的性能优于现有方法.     

结合自适应融合网络与哈希的图像检索算法

卷积神经网络因其对图像识别准确率高而在图像检索领域备受青睐,但处理大规模数据集时,基于卷积神经网络提取的深度特征维度高,容易引发"维度灾难".针对图像检索中深度特征维度高的问题,提出一种基于自适应融合网络特征提取与哈希特征降维的图像检索算法.由于传统哈希处理高维特征复杂度高,因此本文在卷积神经网络中加入自适应融合模块对特征进行重新整合,增强特征表征能力的同时降低特征维度;然后应用稀疏化优化算法对深度特征进行第2次降维,并通过映射获得精简的哈希码;最后,实验以Inception网络作为基础模型,在数据集CIFAR-10和ImageNet上进行了丰富的实验.实验结果表明,该算法能有效提高图像检索效率.     

A robust non-iterative method for similarity transform estimation

A non-iterative and robust method—direct outliers remove (DOR) is proposed, which efficiently estimates the similarity transform based on a data set containing both correct and incorrect correspondences. Unlike hypothesize-and-test methods such as Random Sample Consensus algorithm and its variants, DOR removes mismatches by exploring all the correspondences only once, using two invariant features of similarity transform. One is the angles between two vectors and the other is the length ratios of corresponding vectors. Given two images related by similarity transform, experiments demonstrate that all the mismatches introduced in matching stage could be detected and removed. Without losing computational accuracy, DOR is faster compared with several hypothesize-and-test algorithms, especially when the percentage of correct correspondence is relatively low.     

Learning discriminative representations for semantical crossmodal retrieval

Heterogeneous gap among different modalities emerges as one of the critical issues in multimedia retrieval areas. Unlike traditional unimodal cases, where raw features are extracted and directly measured, the heterogeneous nature of crossmodal tasks requires the intrinsic semantic representation to be compared in a unified framework. Based on a flexible “feature up-lifting and down projecting” mechanism, this paper studies the learning of crossmodal semantic features that can be retrieved across different modalities. Two effective methods are proposed to mine semantic correlations. One is for traditional handcrafted features, and the other is based on deep neural network. We treat them respectively as normal and deep version of our proposed shared discriminative semantic representation learning (SDSRL) framework. We evaluate both of these two methods on two public multimodal datasets for crossmodal and unimodal retrieval tasks. The experimental results demonstrate that our proposed methods outperform the compared baselines and achieve state-of-the-art performance in most scenarios.     

Variational Bayesian inference for infinite generalized inverted Dirichlet mixtures with feature selection and its application to clustering

We developed a variational Bayesian learning framework for the infinite generalized Dirichlet mixture model (i.e. a weighted mixture of Dirichlet process priors based on the generalized inverted Dirichlet distribution) that has proven its capability to model complex multidimensional data. We also integrate a “feature selection” approach to highlight the features that are most informative in order to construct an appropriate model in terms of clustering accuracy. Experiments on synthetic data as well as real data generated from visual scenes and handwritten digits datasets illustrate and validate the proposed approach.     

基于领域本体的贝叶斯网络检索模型研究

针对传统的信息检索方法无法实现用户查询的语义理解、检索效率低等问题,本文提出基于领域本体进行查询扩展的贝叶斯网络检索模型。该模型首先将用户查询通过领域本体进行语义扩展,然后将扩展后的查询作为证据在贝叶斯网络检索模型中进行传播,进而得到查询结果,实验表明本文提出的贝叶斯网络检索模型能提高检索效率。     

一种新型的基于深度学习的单幅图像去雨方法

雨滴严重影响了图像的视觉效果和后续的图像处理应用。目前,基于深度学习的单幅图像去雨方法能够有效挖掘图像的深度特征,其去雨效果优于传统方法;然而,随着网络深度的增加,网络容易出现过拟合的现象,使得去雨效果遇到瓶颈。文中在继承深度学习优点的基础上,学习有雨/无雨图像之间的残差,然后将残差与源图像进行重构,从而获得无雨图像。该方式大幅增加了网络深度,并加快了算法的收敛速度。分别利用通过不同方式获取的雨滴图像对所提方法进行实验验证,并将该方法与当前最新的去雨滴方法作比较,结果表明所提算法的去雨效果更好。     

HAS QoE prediction based on dynamic video features with data mining in LTE network

Evaluation of HTTP adaptive streaming (HAS) quality of experience (QoE) over LTE network is a challenging topic because of multi-segment and multi-rate features of dynamic video sequences. Different from the traditional QoE evaluation methods based on network parameters, this paper proposes the HAS QoE prediction methods based on its dynamic video segment features with data mining. Considering the application requirement of the trade-off between accuracy and complexity, two sets of methodologies are designed to evaluate the HAS QoE including regression and classification. In regression method, we propose the evolved PSNR (ePSNR) model using differential peak signal to noise ratio (dPSNR) statistics as the segment features to evaluate HAS QoE. In classification method, we propose the improved weighted k-nearest neighbors (WkNN) by using dynamic weighted mapping according to the position of video chunk to meet the dynamic segment and rate features of HAS. In order to train and test these methods, we build a real-time HAS video-on-demand (VOD) system in LTE network and do subjective test in different video scenes. With the mean opinion score (MOS), the regression and classification methods are trained to predict the HAS QoE. The validated results show that the proposed ePSNR and WkNN methods outperform other evaluation methods.     

基于特征融合和改进卷积神经网络的环境音识别

由于环境声音复杂的结构,环境声音识别是一个具有挑战性的问题.本文提出一种将特征融合与改进卷积神经网络算法相结合的环境音识别方法.首先针对原始音频文件,提取从波形中学习到的特征以及传统音频特征,分别为MF-CC(梅尔倒谱系数)、GFCC(伽玛通频率倒谱系数)、频谱对比度和CQT(恒定Q变换);然后将提取到的特征分别输入到...     

基于事件的端到端视觉位置识别弱监督网络架构

传统的视觉位置识别（VPR）方法通常使用基于图像帧的相机,存在剧烈光照变化、快速运动等易导致VPR失败的问题。针对上述问题,本文提出了一种使用事件相机的端到端VPR网络,可以在具有挑战性的环境中实现良好的VPR性能。所提出算法的核心思想是,首先采用事件脉冲张量（EST）体素网格对事件流进行表征,然后利用深度残差网络进行...     

PESA-Net: Permutation-Equivariant Split Attention Network for correspondence learning

Establishing reliable correspondences by a deep neural network is an important task in computer vision, and it generally requires permutation-equivariant architecture and rich contextual information. In this paper, we design a Permutation-Equivariant Split Attention Network (called PESA-Net), to gather rich contextual information for the feature matching task. Specifically, we propose a novel “Split–Squeeze–Excitation–Union” (SSEU) module. The SSEU module not only generates multiple paths to exploit the geometrical context of putative correspondences from different aspects, but also adaptively captures channel-wise global information by explicitly modeling the interdependencies between the channels of features. In addition, we further construct a block by fusing the SSEU module, Multi-Layer Perceptron and some normalizations. The proposed PESA-Net is able to effectively infer the probabilities of correspondences being inliers or outliers and simultaneously recover the relative pose by essential matrix. Experimental results demonstrate that the proposed PESA-Net relative surpasses state-of-the-art approaches for pose estimation and outlier rejection on both outdoor scenes and indoor scenes (i.e., YFCC100M and SUN3D). Source codes: https://github.com/x-gb/PESA-Net.     

Combining Shape from Shading and Stereo: A Joint Variational Method for Estimating Depth,Illumination and Albedo

Shape from shading (SfS) and stereo are two fundamentally different strategies for image-based 3-D reconstruction. While approaches for SfS infer the depth solely from pixel intensities, methods for stereo are based on a matching process that establishes correspondences across images. This difference in approaching the reconstruction problem yields complementary advantages that are worthwhile being combined. So far, however, most “joint” approaches are based on an initial stereo mesh that is subsequently refined using shading information. In this paper we follow a completely different approach. We propose a joint variational method that combines both cues within a single minimisation framework. To this end, we fuse a Lambertian SfS approach with a robust stereo model and supplement the resulting energy functional with a detail-preserving anisotropic second-order smoothness term. Moreover, we extend the resulting model in such a way that it jointly estimates depth, albedo and illumination. This in turn makes the approach applicable to objects with non-uniform albedo as well as to scenes with unknown illumination. Experiments for synthetic and real-world images demonstrate the benefits of our combined approach: They not only show that our method is capable of generating very detailed reconstructions, but also that joint approaches are feasible in practice.