- 博客(32)
- 收藏
- 关注
RSS订阅原创 深度学习之实例分割-FCIS
We present the first fully convolutional end-to-end solution for instane-aware semantic segmentation task. It detects and segments the object instances jointly and simultanously. By the introduction
2018-01-21 16:41:54
13142
1
原创 深度学习之实例分割-DeepMask
In this paper, we propose a new way to generate object proposals, introducing an approach based on a discriminative convolutional network. Our model is trained jointly with two objectives: given an
2018-01-21 16:03:54
12190
原创 深度学习之语义分割-PSPNet
In this paper, we exploit the capability of global context information by different-region-based context aggregation through out pyramid pooling module together with the proposed pyramid scene parsing
2018-01-21 15:09:32
3225
原创 深度学习之语义分割-RefineNet
Repeated subsampling operations like pooling or convolution striding in deep CNNs lead to a significant decrease in the initial image resolution. We present RefineNet, a generic multi-path refinemen
2018-01-21 10:30:17
2496
原创 深度学习之语义分割-SegNet
This core trainable segmentation engine consists of an encoder network, a corresponding decoder network followed by a pixel-wise classification layer.模型 说明：基础模型采用VGG16 去掉fc层，使得encoder网络更小，更易训练134M –
2018-01-21 09:25:40
5913
原创 深度学习之语义分割-FCN
We show that convolutional networks by themselves, trained end-to-end, pixels-to-pixels, exceed the state-of-art in semantic segmentation We define a novel architecture that combines semantic inform
2018-01-19 19:58:04
1274
原创 深度学习之检测模型-FPN
Feature pyramids are as basic component in recognition systems for detecting objects at different scales. in this paper, we exploit the inherent multi-scale, pyramids hierarchy of deep convolutiona
2018-01-12 12:52:24
5056
原创 深度学习之实例分割-Mask RCNN
We present a conceptually simple, flexible, and general framework for object instance segmentation. Our approach efficiently detects objects in an image while simultaneously generating a high-quality s
2017-12-12 16:50:28
23345
原创 深度学习之检测模型-Faster RCNN
State-of-the-art object detection networks depend on region proposal algorithms to hypothesize object locations. Advances like SPPnet and Fast R-CNN have reduced the running time of these detection net
2017-12-10 22:55:31
10106
原创 Faster-RCNN检测-RPN
主要贡献提出了 RPN(Region Proposal Networks) 网络来计算候选框 主要步骤:特征提取: 同 Fast R-CNN，以整张图为输入，利用 CNN 得到图像的特征层区域提名: 在最终的卷积特征层conv5-3上利用K 个不同的矩形框 (Anchor Box) 进行提名， k 一般取 9分类与回归: 对每一个 Anchor Box 对应的区域： 进行 object
2017-12-06 13:00:17
1170
原创 深度学习之基础模型-总结
目前来看，很多对 NN 的贡献（特别是核心的贡献），都在于NN的梯度流上，比如- sigmoid会饱和，造成梯度消失。于是有了ReLU。- ReLU负半轴是死区，造成梯度变0。于是有了LeakyReLU，PReLU。- 强调梯度和权值分布的稳定性，由此有了ELU，以及较新的SELU。- 太深了，梯度传不下去，于是有了highway。- 干脆连highway的参数都不要，直接变残差，于是
2017-11-16 17:45:07
2426
1
原创 深度学习之基础模型-SEP-Nets
Albeit there are already intensive studies on compressing the size of CNNs, the considerable drop of performance is still a key concern in many designs. This paper addresses this concern with several n
2017-11-15 23:29:05
1264
原创 深度学习之基础模型-PolyNet
On one hand, the pursuit for very deep networks is met with a diminishing return and increased training difficulty; On the other hand, widening a network would result in a quadratic growth in both comp
2017-11-14 22:58:17
5348
原创 深度学习之基础模型-DenseNet
In this paper, we embrace this observation and introduce the Dense Convolutional Network(DenseNet), which connects each layer to every other layer in a feed-forward fashion. DenseNets have several c
2017-11-05 15:22:30
2843
原创 深度学习之基础模型-ShuffleNet
The new architecture utilizes two proposed operations, point wise group convolution and channel shuffle, to greatly reduce computation cost while maintaining accuracy.该论文利用了point wise group convolutio
2017-11-04 23:59:53
1499
原创 深度学习之基础模型-FractalNet
We introduce a design strategy for neural network macro-architecture based on self-similarity. Repeated application of a simple expansion rule generates deep networks whose structural layouts are preci
2017-11-01 00:14:42
2823
1
原创 深度学习之基础模型-Xception
We present an interpretation of Inception modules in convolutional neural networks as being an intermediate step in-between regular convolution and the depth wise separable convolution operation(a dept
2017-10-30 18:08:44
18027
原创 深度学习之基础模型-mobileNet
MobileNets are based on streamlined architecture that uses depthwise separable convolutions to build light weight deep nueral network. 思想基于depthwise separable convolution(来源于FactorizedNet)来实现 depth
2017-10-24 21:40:14
7738
2
原创 深度学习之基础网络-SqueezeNet
For a given accuracy level, it is typically possible to identify multiple CNN architecture that achieve that accuracy level.思想现在研究深度卷机模型，主要有两个出发点：给定数据集，提高精度给定精度，降低模型参数大小参数少的好处：在分布式训练更有效：more effic
2017-10-23 20:42:26
1481
原创 深度学习之基础网络-Inception-V4
One example is the Inception architecture that has been shown to achieve very good performance at relatively low computational cost. Recently, the introduction of residual connections in conjunction wi
2017-10-21 19:36:33
1438
原创 深度学习之基础模型-Inception-V3
Although increased model size and computational cost tend to translate to immediate quality gains for most tasks (as long as enough labeled data is provided for training), computational efficiency and
2017-10-19 20:59:52
13343
原创 深度学习之基础模型-Inception-V2（BN）
Training Deep Neural Networks is complicated by the fact that the distribution os each layer’s inputs changes during training, as the parameters of the previous layers change作者指出，深度网络的一直都非常困难，主要是由于在训练
2017-10-17 12:58:35
4019
原创 深度学习之基础模型—resNet
The depth of representation is of central importance for many visual recognition tasks继VGG和GoogLeNet在网络深度上进行了进一步的尝试，取得了比较大的进展-网络越深效果也越好-，但也遇到了问题： 网络越深，越容易出现梯度消失，导致模型训练难度变大，出现“退化”现象注：从VGG的结果也能感
2017-10-13 18:12:21
14281
1
原创 深度学习之基础模型—InceptionV1(GoogLeNet)
与VGG同年，google也独立在深度网络的深度方向进行了研究。发明了一个22层的深度神经网络，为了致敬LeNet，取名GoogLeNet（非GoogleNet）。该结构充分利用了网络结构中子网络的计算，使得网络的深度进一步提升。在设计过程中，受到了多尺度处理的启发，并且结合Hebbian principle原理进行网络结构设计，使得网络达到最优的状态。简介GoogLeNet的参数量相比AlexN
2017-10-05 08:25:33
6324
原创 深度学习之基础模型-VGG
网络结构结构Architecture模型框架效果分析单尺度多尺度多尺度裁剪模型融合对比总结参考文献VGG论文给出了一个非常振奋人心的结论：卷积神经网络的深度增加和小卷积核的使用对网络的最终分类识别效果有很大的作用。记得在AlexNet论文中，也做了最后指出了网络深度的对最终的分类结果有很大的作用。这篇论文则更加直接的论证了这一结论。作者Karen Simonyan & Andre
2017-09-27 12:45:43
32366
6
原创 深度学习之基础模型-NIN
Network in Network 是google提出的一种在网络结构上非常具有创造性的文章。以前的深度学习结构，往往是CNN+FC之类的组合，这样在CNN与FC进行结合的过程中，则把两部分过程独立来开，CNN负责提取特征，FC负责特征分类。而NIN则利用mlpconv和global average pooling将CNN和FC两部分有机的结合在一起，使得其可解释性更强。从而文章的创新点： *提出
2017-09-24 10:18:31
2496
原创 深度学习之基础模型---AlexNet
论文发表于2012年，由深度学习之父Hinton和他的学生发表。Alex Krizhevshy, Ilya Sutskever Geoffre E.Hinton。Hinton在深度学习领域的地位不用多说，神级别的人物，本人只能跪拜了。摘要论文训练一个非常大，深的卷机神经网络，针对120w的高精度图像（来自ImageNet LSVRC-2010）进行1000类进行识别。在测试集中，top1和top5的
2017-09-13 14:36:52
1873
原创 tensorflow学习之卷积
[Toc] tensorflow中提供了多种卷机操作，如下tf.nn.conv2dtf.nn.depthwise_conv2dtf.nn.separable_conv2dtf.nn.atrous_conv2d下面我们逐一对卷机进行分析Tensortensorflow中的图像与卷机核都是一个4维tensor，tensor的每一位表达的含义如下 inputdata=[batchnum,in
2017-09-12 17:31:04
520
原创 卷积与反卷积
卷机与反卷机卷机(Convolution)卷机定义：图像中的卷机，可以理解为在空间位置上，进行加权叠加。 ——[图片来源]假设输入特征层feature map的大小为[batch,width,height,channel][batch, width,height,channel]， 卷积核大小[k,k,inputn,outputn][k,k,input_n,output_n]，步长为[1,s,
2017-09-11 16:32:06
926
原创 文章标题
＃图像处理中的频率分析在图像处理中，我们经常会用把图像转化到频率进行处理，最常用的就是傅里叶变换。把图像转化成频率后，频率中的坐标怎么和图像中的原始信息对应呢？例如，我们想去掉图像中的纹理信息[原始图像](http://img.blog.youkuaiyun.com/2016101219195960
2016-10-12 19:21:12
245
转载 欢迎使用优快云-markdown编辑器
欢迎使用Markdown编辑器写博客本Markdown编辑器使用StackEdit修改而来，用它写博客，将会带来全新的体验哦：Markdown和扩展Markdown简洁的语法代码块高亮图片链接和图片上传LaTex数学公式UML序列图和流程图离线写博客导入导出Markdown文件丰富的快捷键快捷键加粗 Ctrl + B 斜体 Ctrl + I 引用 Ctrl
2016-10-11 20:52:40
222
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