- 博客(42)
- 收藏
- 关注
RSS订阅
原创 pytorch: A 60 Minute blitz笔记
A replacement for NumPy to use the power of GPUsa deep learning research platform that provides maximum flexibility and speed1. Basic1.1 tensorsx = torch.empty(5, 3) # unitializedx = torch...
2018-05-14 14:29:12
504
原创 爬虫的一些经验和思考
微博爬虫的一些经验和思考微博爬虫的一些经验和思考1. 总结2. todos3. 基本路线4. cookie5. 代理6. 验证码7. Others1. 总结如果量不大，大型网站(微博，知乎)，建议直接使用路线2，省时省力 如果量不大，小型网站(反爬策略弱)，直接路线12. todos分布式爬虫 自建框架scrapy抓包，...
2018-05-07 00:29:29
499
原创 图片处理常用函数和代码
numpyjitterclippreprocessndarray[None]numpyjitter# Randomly jitter the image a bitox, oy = np.random.randint(-max_jitter, max_jitter+1, 2)X = np.roll(np.roll(X, ox, ...
2018-04-23 20:28:59
279
原创 tensorflow 问题记录
计算梯度tf.gradients(ys, xs) symbolic derivatives of sum of ys w.r.t x in xsys, xs tensorreturn: A list of sum(dy/dx) for each x in xs注意，返回的是y整个tensor对x每个元素的梯度，也即返回和x的shape相同cross-entropy loss...
2018-04-20 15:48:34
235
原创 ReadingList
优先Few-shot Autoregressive Density Estimation: Towards Learning to Learn DistributionsMAMLSummary MAML: meta-learn an initial conditionLSTM optimization: meta-learn a good initial cond...
2018-04-15 18:36:34
618
原创 tensorflow基础知识
1. Graph operators1.1 collection常用数值计算conditionLOW LEVEL API1. Introduction1.1 data -> tensor1.2 computation1.3 Session1.4 Feeding1.5 layers2. Tensor2.1 shape2....
2018-01-27 17:42:04
583
原创 Context Encoding for Semantic Segmentation[CVPR2018]
FCN frameworkGood explanationglobal receptive fields: conv(no linearities) + downsamplespatial resolution lossencoder: dilated conv pro: expand receptive fieldcon: isolate pixels from cont...
2018-09-04 22:36:43
901
原创 深度学习训练常用参数含义解释
momentum: μμ\mu,上一次更新的权重(保证更新方向不会和上次偏离过多) learning rate: αα\alpha,当前梯度的权重 weight decay: λλ\lambda 相当于L2 惩罚 refs: caffe,pytorch,stackoverflow Vt+1=μVt−α∇L(Wt)Vt+1=μVt−α∇L(Wt)V_{t+1} = \mu V_t - \a...
2018-08-31 22:01:52
1769
原创 Context Encoders: Feature Learning by Inpainting 阅读笔记
Github目录1. 介绍2. 架构3. 数据生成1. 介绍无监督学习，图像修复学习深层特征通过下图可以看到效果很好 2. 架构 link 采用类似encoder-decoder架构, L2 loss使图片重建相近，adversarial loss使图片清晰 encoder和decoder之间使用channel-wis...
2018-08-31 13:02:31
8675
3
原创 Adversarial Learning for Semi-Supervised Semantic Segmentation(BMVC2018)
Adversarial Learning for Semi-Supervised Semantic Segmentation本文核心要点如下SegNet as GeneratorDiscriminator: use a FCN(HxWxC -> HxWx1)Semi-supervised: 对于unlabeled data, adv loss, (fix D, maximi...
2018-08-04 17:06:28
1015
原创 Image-level to Pixel-wise Labeling: From Theory to Practice[IJCAI18]
Image-level to Pixel-wise Labeling: From Theory to Practiceincorporate image label in segmentation networkuse generative network to recover original image and constrainanalyze effects of image la...
2018-07-29 19:00:19
382
原创 pytorch 问题记录
Tensor & VariableWhat is the difference between Tensors and Variables in Pytorch?torch tensors are actually the data.variables wrap tensors, and construct a chain of operations between the t...
2018-05-24 17:04:55
1959
原创 Deeplabv3+ 阅读笔记
Notes: 谷歌deeplabv3+的代码现在已经开源，详见deeplab(Github)，还有一个使用的demo样例。0.spatial pyramid poolingprobing the incoming features with filters or pooling operations at multiple rates and multiple effecti...
2018-04-25 21:31:41
6817
原创 Zero-Shot Visual Recognition using Semantics-Preserving Adversarial Embedding Networks 阅读笔记
Zero-Shot Visual Recognition using Semantics-Preserving Adversarial Embedding Networks收获多个网络如何组合? 相同目标collaborative, 不同目标adversarialtodo查看groud truth class embedding查看SAEConclusio...
2018-04-17 20:21:11
700
原创 Transductive Unbiased Embedding for Zero-Shot Learning阅读笔记
Transductive Unbiased Embedding for Zero-Shot LearningSummaryPROubias term: 在Loss添加一个针对未知类的loss, 部分抑制了zero shot天生倾向于带label数据的问题巧妙的数据利用，虽然target dataset没有用label(图片文字对应关系)，但是用了label的文字embedd...
2018-04-11 13:39:45
1338
2
原创 Object Region Mining with Adversarial Erasing: A Simple Classification to Semantic Segmentation阅读笔记
Object Region Mining with Adversarial Erasing: A Simple Classification to Semantic Segmentation Approach阅读笔记x. QuestionsAE步骤, 怎么判断收敛不好。万一有的图片需要erase 3次，有的erase2两次会怎么样? 实验直接选了3次 PSL，为什么conv7的结...
2018-04-01 21:42:51
1455
原创 图像分割(不断更新)
弱监督语义分割what: segment image only use image-level labelchallenge: how to build relation between image labels and pixels 参考1. Features针对输入图片的salience map gradient based针对最后一层的class activa...
2018-04-01 21:41:11
286
原创 cs231n 12 Visualizing and Understanding
Visualizing and Understandingwhat’s going on in CNNFirst layer:weights, Filter, visual layers because when input similar to weights, result will be maximizedhigher layer filters: meaning less...
2018-04-01 13:15:30
232
原创 A 2017 Guide to Semantic Segmentation with Deep Learning 笔记
原文A 2017 Guide to Semantic Segmentation with Deep Learning0. Intro1. Problem1.1 before deep1.2 current1.3 postprocessing2. Models0. Intromainly use natural/real world im...
2018-03-23 17:26:37
276
原创 cs231n lecture13 Generative Models
Generative ModelsPixelCNN/RNNVAEGAN PixelRNN/CNN生成速度慢pro Can explicitly compute likelihood p(x)Explicit likelihood of training data gives good evaluation metricGood samplescon Sequ...
2018-03-23 16:26:07
354
原创 cs231n lecture11 segmentation
Segmentation, Localization, DetectionSemantic Segmentationlabel each pixel in the image with a category labelknow classesidea: sliding window inefficient, Not reusing shared features between...
2018-03-21 15:29:41
279
原创 cs231n lecture9 CNN Architectures
AlexNetCONV - MAXPOOL - NORM(not common) CONV - MAXPOOL - NORM CONV - CONV - CONV - MAXPOOL FC - FC - FC local response normalizationVGGdeeper (16-19layers)small filter (3x3 CONV stride...
2018-03-20 16:56:26
223
原创 cs231n lecture5 CNN
CNN 笔记Convolution LayerPooling LayerFully Connected Layer(FC layer)Useful NotesPreprocessingWeigth InitializationRegularizationLossclassificationAttribute classificationregression...
2018-03-20 14:47:45
192
原创 mysql 批量kill
mysql> select concat('KILL ',id,';') from information_schema.processlist where user='root';+------------------------+| concat('KILL ',id,';') |+------------------------+| KILL 3101; ...
2018-03-13 13:32:48
967
原创 网络路由设置
route -nroute add $host $gwip dev em1 route add defaut gw $gwiproute 命令添加的路由，机器重启或者网卡重启后就没掉了，在linux下设置永久路由的方法：1.在/etc/rc.local里添加 2.在/etc/sysconfig/network里添加到末尾 3./etc/sysconfig/static-route...
2018-03-09 16:20:02
169
原创 NN.lecture 2
algorithmsmemory-basederror-correcthebbian postulatecompetetiveparadigmssupervisedunsupervisedreinforcement机器学习理论:训练数据集与测试数据集具有相同分布的假设 解决实际问题，很多不成立...
2018-03-08 20:33:25
189
原创 Matching Networks for one Shot Learning 阅读笔记
Conclusionparametric vs metric karpathy很好的paper notesAdvantages使用加权平均(metric)方法去做few-shot,优点: 速度快，需要训练量小将对S的训练加入模型 pθ(y|x,S)pθ(y|x,S)p_\theta(y|x,S)，具体做法是:分成task(或称为训练集)去训练, 每次选择一些类别L(label s...
2018-03-07 17:50:10
10964
1
原创 cs224n lecture8 Recurrent Neural Networks and Language Models
Traditional language modelsRNNsRNN language modelstraining problems and tricksRNN for other sequence tasksBi and deep rnnsLanguage Modelscomputes a probability for a sequence of words P(...
2018-03-06 23:08:14
242
原创 Learning to learn
字体数据集 imagenet几个应用few shot classification (image) + RL + 推荐(cold start)
2018-03-05 21:57:12
426
原创 DL学习笔记16 Structured Probabilistic Models for Deep Learning
Structured Probabilistic Models阅读笔记原文link, slides笔记0. IntroStructured probabilistic model:a way of describing a probability distritutionalso called graphic modelsRoadmapchallege...
2018-02-25 23:36:19
657
原创 python离线安装包
1. 生成依赖pip freeze > requirements.txt2. 下载包pip download -r requirements.txt ref3. 离线安装包pip install --no-index --find-links /path/to/some/dir/ celery pip install --no-index --find-lin...
2018-02-13 16:17:55
669
原创 Linux 使用sshpass和pssh集群批量安装
pssh 可以实现多台机器并行安装 sshpass 可以传递密码,针对没有免密的情况 参考ssh-cluster-administration sshpass -f remotePasswordFile parallel-ssh -I -A -h hostlist "sudo -S apt-get dist-upgrade" < remoteRootPasswordFile “...
2018-02-13 12:41:05
935
原创 conda使用
create virtualenv conda create --name py2.7 python=2.7 activate source activate env 待续
2018-02-01 20:30:50
164
转载 select into from 和 insert into select
数据库表复制转自select into from 和 insert into select 的用法和区别 select into from 和 insert into select都是用来复制表，两者的主要区别为： select into from 要求目标表不存在，因为在插入时会自动创建。insert into select from 要求目标表存在备份表数据： create ta
2018-02-01 09:53:59
148
转载 sigmoid和softmax的关系
softmax P(y=k)=exp(wkT⋅x)∑kexp(wkT⋅x)" role="presentation">P(y=k)=exp(wTk⋅x)∑kexp(wTk⋅x)P(y=k)=exp(wkT⋅x)∑kexp(wkT⋅x)P(y=k) = \frac{exp(\mathbf w_k^T \cdot \mathbf x)}{\sum_{k}e
2018-01-29 22:56:28
1020
转载 The secret ingredients of word2vec 概要整理
源自Sebastian Ruder的博文The secret ingredients of word2vecQuestionsQ1. Are embeddings superior to distributional methods?With the right hyperparameters, no approach has a consistent advantage over
2018-01-20 23:08:26
186
原创 RTMP文件下载
RTMP文件下载今天在使用techtalks观看ACL视频时，比较卡顿，萌生了下载的想法查看一下网页源代码,是flash播放，使用了flowplayerRTMP文件下载下载1 依赖2 下载相关知识1 TECHTALK2 flowplayer21 简介22 RTMP23 url24 clip25 key3 FLASH URL
2018-01-16 22:12:25
5016
原创 Centos Yum 配置&&常用软件安装
Yum 配置指南centos6.6修改默认Yum源# 使用阿里镜像wget -O /etc/yum.repos.d/CentOS-Base.repo http://mirrors.aliyun.com/repo/Centos-6.repo# 生成缓存yum makecache安装EPEL, IUS repositorysudo yum install epel-r
2018-01-15 11:35:31
383
原创 垃圾评论检测
Opinion Spam DetectionProblemorigin: Positive opinions often mean profits and fames for businesses and individualslead to: promote or to discredit some target products, services, organizations, indi
2018-01-10 22:40:44
3193
2
转载 Reinforcement Learning for Relation Classification from Noisy Data阅读笔记
Reinforcement Learning for Relation Classification from Noisy Data阅读笔记
2017-11-30 23:02:53
2237
空空如也
空空如也
TA创建的收藏夹 TA关注的收藏夹
TA关注的人