【Paper Reading】Deep Supervised Hashing for fast Image Retrieval
what has been done:
This paper proposed a novel Deep Supervised Hashing method to learn a compact similarity-presevering binary code for the huge body of image data.
Data sets:
CIFAR-10: 60,000 32*32 belonging to 10 mutually exclusively categories(6000 images per category)
NUS-WIDE: 269,648 from Flickr, warpped to 64*64
content based image retrieval: visually similar or semantically similar.
Traditional method: calculate the distance between the query image and the database images.
Problem: time and memory
Solution: hashing methods(map image to compact binary codes that approximately preserve the data structure in the original space)
Problem: performace depends on the features used, more suitable for dealing with the visiual similarity search rather than the sematically similarity search.
Solution: CNNs, the CNNs successful applications of CNNs in various tasks imply that the feature learned by CNNs can well capture the underlying sematic structure of images in spite of significant appearance variations.
Related works:
Locality Sensitive Hashing(LSH):use random projections to produce hashing bits
cons: requires long codes to achieve satisfactory performance.(large memory)
data-dependent hashing methods: unsupervised vs supervised
unsupervised methods: only make use of unlabelled training data to lean hash functions
- spectral hashing(SH): minimizes the weighted hamming distance of image pairs
- Iterative Quantization(ITQ): minimize the quantization error on projected image descriptors so as to allievate the information loss
supervised methods: take advantage of label inforamtion thus can preserve semantic similarity
- CCA-ITQ: an extension of iterative quantization
- predictable discriminative binary code: looks for hypeplanes that seperate categories with large margin as hash function.
- Minimal Loss Hashing(MLH): optimize upper bound of a hinge-like loss to learn the hash functions
problem: the above methods use linear projection as hash functions and can only deal with linearly seperable data.
solution: supervised hashing with kernels(KSH) and Binary Reconstructive Embedding(BRE).
Deep hashing: exploits a non-linear deep networks to produce binary code.
Problem : most hash methods relax the binary codes to real-values in optimizations and quantize the model outputs to produce binary codes. However there is no guarantee that the optimal real-valued codes are still optimal after quantization .
Solution: DIscrete Graph Hashing(DGH) and Supervided Discrete Hashing(DSH) are proposed to directly optimize the binary codes.
Problem : Use hand crafted feature and cannot capture the semantic information.
Solution: CNNs base hashing method
Our goal: similar images should be encoded to similar binary codes and the binary codes should be computed efficiently.
Loss function:
Relaxation:
Implementation details:
Network structure:
3*卷积层:
3*池化层:
2*全连接层:
Training methodology:
- generate images pairs online by exploiting all the image pairs in each mini-batch. Allivate the need to store the whole pair-wise similarity matrix, thus being scalable to large-scale data-sets.
- Fine-tune vs Train from scratch
Experiment:
CIFAR-10
GIST descriptors for conventional hashing methods
NUS-WIDE
225-D normalized block-wise color moment features
Evalutaion Metrics
mAP: mean Average Precision
precision-recall curves(48-bit)
mean precision within Hamming radius 2 for different code lengths
Network ensembles?
Comparison with state-of-the-art method
CNNH: trainin the model to fit pre-computed discriminative binary code. binary code generation and the network learning are isolated
CLBHC: train the model with a binary-line hidden layer as features for classification, encoding dissimilar images to similar binary code would not be punished.
DNNH: used triplet-based constraints to describe more complex semantic relations, training its networks become more diffucult due to the sigmoid non-linearlity and the parameterized piece-wise threshold function used in the output layer.
Combine binary code generation with network learning
Comparision of Encoding Time
【Paper Reading】Deep Supervised Hashing for fast Image Retrieval的更多相关文章
- 【Paper Reading】Learning while Reading
Learning while Reading 不限于具体的书,只限于知识的宽度 这个系列集合了一周所学所看的精华,它们往往来自不只一本书 我们之所以将自然界分类,组织成各种概念,并按其分类,主要是因为 ...
- 【Paper Reading】Object Recognition from Scale-Invariant Features
Paper: Object Recognition from Scale-Invariant Features Sorce: http://www.cs.ubc.ca/~lowe/papers/icc ...
- 【Paper Reading】Bayesian Face Sketch Synthesis
Contribution: 1) Systematic interpretation to existing face sketch synthesis methods. 2) Bayesian fa ...
- 【Paper Reading】Improved Textured Networks: Maximizing quality and diversity in Feed-Forward Stylization and Texture Synthesis
Improved Textured Networks: Maximizing quality and diversity in Feed-Forward Stylization and Texture ...
- 【资料总结】| Deep Reinforcement Learning 深度强化学习
在机器学习中,我们经常会分类为有监督学习和无监督学习,但是尝尝会忽略一个重要的分支,强化学习.有监督学习和无监督学习非常好去区分,学习的目标,有无标签等都是区分标准.如果说监督学习的目标是预测,那么强 ...
- 【文献阅读】Deep Residual Learning for Image Recognition--CVPR--2016
最近准备用Resnet来解决问题,于是重读Resnet的paper <Deep Residual Learning for Image Recognition>, 这是何恺明在2016-C ...
- 【文献阅读】Augmenting Supervised Neural Networks with Unsupervised Objectives-ICML-2016
一.Abstract 从近期对unsupervised learning 的研究得到启发,在large-scale setting 上,本文把unsupervised learning 与superv ...
- 【CS-4476-project 6】Deep Learning
AlexNet / VGG-F network visualized by mNeuron. Project 6: Deep LearningIntroduction to Computer Visi ...
- 【论文阅读】Deep Mixture of Diverse Experts for Large-Scale Visual Recognition
导读: 本文为论文<Deep Mixture of Diverse Experts for Large-Scale Visual Recognition>的阅读总结.目的是做大规模图像分类 ...
随机推荐
- vue 截取视频第一帧
最近自己写项目,在项目中涉及功能点又截取视频帧的点:需求澄清:移动端封面展示,平台上传图片(多张上传)取第一张上传图片为封面图:如上传视频则截取视频第一帧作为封面图: 实现思路:h5 video标签 ...
- VUE 利用 webpack 给生产环境和发布环境配置不同的接口地址
转载地址: https://blog.csdn.net/gebitan505/article/details/58166055 VUE 利用 webpack 给生产环境和发布环境配置不同的接口地址 前 ...
- 使用maven创建springMVC时返回页面报错
这是由于你的 Maven 编译级别是 jdk1.5 或以下,而你导入了 jdk1.6 以上的依赖包 解决办法: <build> <finalName></finalNam ...
- 《黑白团团队》第八次团队作业:Alpha冲刺 第一天
项目 内容 作业课程地址 任课教师首页链接 作业要求 团队项目 填写团队名称 黑白团团队 填写具体目标 认真负责,完成项目 团队项目Github仓库地址链接. [Alpha] Scrum meetin ...
- socket 客户端的认证
一:使用 hashlib 进行加密验证: # server.py 服务端 import os import socket import hashlib def check_conn(conn): ...
- BA-风阀水阀执行器接线图
220水阀执行器接线图 24V风阀执行器接线图
- Linux线程资源限制
- mysql安装出错cannot create windows service for mysql.error:0
配置时最后一步出现不能启动mysql 解决成功的办法:[MySQL] Could not start the service MySQL 解决方法 安装mysql 5.1.33,在运行Server I ...
- word2010无法显示endnote x7插件及破解endnote x7
最近本人由于要写文章需要使用endnotex7,相比于mendeley和noteexpress,文献管理和引用我喜欢endnote x7,阅读喜欢mendeley.可是由于之前用的正版30天到期了,破 ...
- springMVC+uploadify3.1 文件上传 demo
uploadify3.1 api 可参考:(点击打开链接) 需要springmvc的jar包 1.upload.jsp(主要代码) <script type="text/javascr ...