论文笔记
文章平均质量分 95
不甘心的程序员
这个作者很懒,什么都没留下…
展开
-
论文笔记 - Cross-dimensionalWeighting for Aggregated Deep Convolutional Features
abstractIntroductionRelated workFramework for Aggregation of Convolutional FeaturesFramework OverviewCross-dimensional WeightingFeature Weighting SchemesResponse Aggregation for Spatial We...翻译 2018-05-10 10:24:14 · 2737 阅读 · 4 评论 -
论文笔记 - kernel-based non-parametric activation functions for neural networks
abstractintroductionpreliminariesFixed activation functionsParametric adaptable activation functionsNon-parametric activation functionsAPL functions 分段线性插值函数spline functions 样条插值 (SAF)maxou...翻译 2018-04-26 12:09:00 · 414 阅读 · 0 评论 -
论文笔记 - Deep Semi-Random Features for Nonlinear Function Approximation
AbstractIntroductionBackgroundSemi-Random FeaturesOne Hidden Layer ModelUniversal Approximation AbilityOptimization TheoryGeneralization GuaranteeMultilayer ModelBenefit of DepthOptimiz...翻译 2018-04-26 19:30:01 · 252 阅读 · 0 评论 -
论文笔记 - NetVLAD: CNN architecture for weakly supervised place recognition
abstractintroductionmethod overviewDeep architecture for place recognitionNetVLAD: A Generalized VLAD layer (fVLADfVLAD f_{VLAD})Max pooling (fmax)Learning from Time Machine dataExperiment...翻译 2018-04-27 20:30:35 · 10023 阅读 · 0 评论 -
论文笔记 - Learning Compact Binary Descriptors with Unsupervised Deep Neural Networks
abstractIntroductionRelated WorkBinary DescriptorsApproachOverall Learning ObjectivesLearning Discriminative Binary DescriptorsLearning Efficient Binary DescriptorsLearning Rotation Invar...翻译 2018-05-06 20:30:18 · 1733 阅读 · 0 评论 -
论文笔记 - Multi-Scale Orderless Pooling of Deep Convolutional Activation Features
AbstractIntroductionThe Proposed MethodAnalysis of InvarianceLarge-Scale EvaluationDatasetsImage Classication ResultsImage Retrieval ResultsAbstractCNN在图像识别应用上性能很好,但是全局CNN特征缺少几何...翻译 2018-05-19 15:43:19 · 1074 阅读 · 0 评论 -
论文笔记 - Weighted Component Hashing of Binary Aggregated Descriptors for Fast Visual Search
AbstractIntroductionWeCoHash image search frameworkRelate WorksAggregated DescriptorsBinary Aggregated DescriptorsHashing Binary Vectors for ANN SearchAbstract本文要解决的问题:尽管计算汉明距离速...翻译 2018-05-15 10:54:20 · 281 阅读 · 0 评论 -
[论文笔记] - PARTICULAR OBJECT RETRIEVAL WITH INTEGRAL MAX-POOLING OF CNN ACTIVATIONS
AbstractAbstractCNN能够提取出有效的特征描述子来表示图像,但是对于compatible with geometry-aware re-ranking 几何感知重排序方法(兼容??) 本文重新回顾了检索过程,初始搜索和重排序 构建了紧致的特征向量,编码图像区域,不需要重复将图像送入网络中。 利用积分图像来最大pooling bound ing b...翻译 2018-05-26 16:58:40 · 3481 阅读 · 0 评论 -
DMHI
DMIH基于object的图像检索,对多尺度的instance进行hash编码 解决的问题是:通常在图像检索中,使用一张图像,找到与其相类似的图像,本文的想法是:通过输入多张图像来检索同时包含这多张图像中物体的图像。比如输入奥巴马和普京的照片,检索得到奥巴马和普京共同存在的图像。 本文的方法:使用一种弱监督的DMIH框架来实现图像检索,这个框架包含两部分内容,物体的检测以及对物体的hash...翻译 2018-06-28 00:48:22 · 681 阅读 · 2 评论