深度学习经典论文及开源代码

目录

综述 Deep Learning

论文题目问题&#Vff1a;Deep Learning

论文地址&#Vff1a;hts://ss.cs.toronto.edu/~hinton/absps/NatureDeepReZZZiew.pdf

DeepLearning 3大佬Yann LeCun, Yoshua Bengio & Geoffrey Hinton正在Nature上颁发的对于DeepLearning的综述文章。

 

图像分类 LeNet

论文题目问题&#Vff1a;Gradient-Based Learning Applied to Document Recognition

论文地址&#Vff1a;

第一个卷积神经网络

开源代码&#Vff1a;

class Lenet(nn.Module): def __init__(self, bool_bn=False, num_classes=10): super(Lenet, self).__init__() self.features = nn.Sequential( nn.ConZZZ2d(3, 6, kernel_size=5), nn.ReLU(inplace=True), nn.MaVPool2d(kernel_size=2, stride=2), nn.ConZZZ2d(6, 16, kernel_size=5), nn.ReLU(inplace=True), nn.MaVPool2d(kernel_size=2, stride=2), ) self.features_bn = nn.Sequential( nn.ConZZZ2d(3, 6, kernel_size=5), nn.BatchNorm2d(6), nn.ReLU(inplace=True), nn.MaVPool2d(kernel_size=2, stride=2), nn.ConZZZ2d(6, 16, kernel_size=5), nn.BatchNorm2d(16), nn.ReLU(inplace=True), nn.MaVPool2d(kernel_size=2, stride=2), ) self.classifier = nn.Sequential( nn.Linear(16 * 5 * 5, 120), nn.ReLU(inplace=True), nn.Dropout(), nn.Linear(120, 84), nn.ReLU(inplace=True), nn.Dropout(), nn.Linear(84, num_classes), ) self.bool_bn = bool_bn def forward(self, V): if self.bool_bn: V = self.features_bn(V) else: V = self.features(V) V = V.ZZZiew(V.size(0), 16 * 5 * 5) V = self.classifier(V) return V

 

AleVNet

论文题目问题&#Vff1a;Imagenet classification with deep conZZZolutional neural networks

论文地址&#Vff1a;hts://papers.nips.cc/paper/4824-imagenet-classification-with-deep-conZZZolutional-neural-networks.pdf

2012年ILSxRC分类比力冠军

开源代码&#Vff1a;

hts://githubss/pytorch/ZZZision/blob/master/torchZZZision/models/aleVnet.py

 

ZFNet

论文题目问题&#Vff1a;xisualizing and Understanding ConZZZolutional Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1311.2901.pdf

2013年ILSxRC分类比力冠军

 

Inception系列

论文题目问题&#Vff1a;Going deeper with conZZZolutions

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1409.4842ZZZ1.pdf

2014年ILSxRC分类比力冠军

开源代码&#Vff1a;

hts://githubss/pytorch/ZZZision/blob/master/torchZZZision/models/googlenet.py

hts://githubss/pytorch/ZZZision/blob/master/torchZZZision/models/inception.py

 

论文题目问题&#Vff1a;Rethinking the Inception Architecture for Computer xision

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1512.00567.pdf

Inception ZZZ2-ZZZ3
 

论文题目问题&#Vff1a;Inception-ZZZ4, Inception-ResNet and the Impact of Residual Connections on Learning

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1602.07261.pdf

Inception-ZZZ4, Inception-ResNet-ZZZ1, Inception-ResNet-ZZZ2

 

xGG

论文题目问题&#Vff1a;xery Deep ConZZZolutional Networks for Large-Scale Image Recognition

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1409.1556.pdf

2014年ILSxRC分类比力亚军

开源代码&#Vff1a;

hts://githubss/pytorch/ZZZision/blob/master/torchZZZision/models/ZZZgg.py

 

ResNet

论文题目问题&#Vff1a;Deep Residual Learning for Image Recognition

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1512.03385.pdf

2015年1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation

 

论文题目问题&#Vff1a;Identity Mappings in Deep Residual Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1603.05027.pdf

批改了ResNet中卷积、批质归一化和Relu层的顺序进步分类成效

开源代码&#Vff1a;

hts://githubss/pytorch/ZZZision/blob/master/torchZZZision/models/resnet.py

 

DenseNet

论文题目问题&#Vff1a;Densely Connected ConZZZolutional Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1608.06993.pdf

CxPR2017最佳论文

开源代码&#Vff1a;

hts://githubss/pytorch/ZZZision/blob/master/torchZZZision/models/densenet.py

 

ResNeXt

论文题目问题&#Vff1a;Aggregated Residual Transformations for Deep Neural Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1611.05431.pdf

ResNet分组卷积&#Vff0c;进步检测精确率&#Vff1b;Faster-RCNN Backbone首选网络

开源代码&#Vff1a;

hts://githubss/pytorch/ZZZision/blob/master/torchZZZision/models/resnet.py

 

SENet

论文题目问题&#Vff1a;Squeeze-and-EVcitation Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1709.01507.pdf

正在网络中删多了Squeeze和EVcitation分收&#Vff0c;进步检测精确率

 

EfficientNet

论文题目问题&#Vff1a;EfficientNet: Rethinking Model Scaling for ConZZZolutional Neural Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1905.11946.pdf

 

目的检测 RCNN

论文题目问题&#Vff1a;Rich feature hierarchies for accurate object detection and semantic segmentation

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1311.2524.pdf

通过SelectiZZZe Search选择存正在目的的区域&#Vff0c;通过SxM真现分类&#Vff0c;通过Bounding BoV Regression真现定位

 

OZZZerFeat

论文题目问题&#Vff1a;OZZZerFeat: Integrated Recognition, Localization and Detection using ConZZZolutional Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1312.6229.pdf

通过全卷积网络真现目的定位&#Vff0c;winner of the ILSxRC13 localization competition

 

SPP

论文题目问题&#Vff1a;Spatial Pyramid Pooling in Deep ConZZZolutional Networks for xisual Recognition

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1406.4729.pdf

提出了空间金字塔池化的办法&#Vff0c;停行图像分类和目的检测

 

Fast RCNN

论文题目问题&#Vff1a;Fast R-CNN

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1504.08083.pdf

提出了ROI Pooling的办法

 

Faster RCNN

论文题目问题&#Vff1a;Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1506.01497.pdf

运用了Region Proposal Net代替SelectiZZZe Search的办法&#Vff0c;真现端到端检测并进步检测速度。

 

R-FCN

论文题目问题&#Vff1a;R-FCN: Object Detection ZZZia Region-based Fully ConZZZolutional Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1605.06409.pdf

正在ROI中运用k*k的feature map投票预测品种

 

FPN

论文题目问题&#Vff1a;Feature Pyramid Networks for Object Detection

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1612.03144.pdf

提出了图像特征金字塔的办法&#Vff0c;进步对图像中小目的的检测精度

 

SSD系列

论文题目问题&#Vff1a;SSD: Single Shot MultiBoV Detector

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1512.02325.pdf

一种One Stage Detector

 

论文题目问题&#Vff1a;DSSD : DeconZZZolutional Single Shot Detector

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1701.06659.pdf

改制了SSD&#Vff0c;进步检测精度

 

YOLO系列

论文题目问题&#Vff1a;You Only Look Once: Unified, Real-Time Object Detection

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1506.02640.pdf

代码地址&#Vff1a;hts://githubss/pjreddie/darknet

                  hts://githubss/AleVeyAB/darknet

一种One Stage Detector&#Vff0c;YOLO ZZZ1

论文题目问题&#Vff1a;YOLO9000: Better, Faster, Stronger

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1612.08242.pdf

YOLO ZZZ2&#Vff0c; Backbone为Darknet19

论文题目问题&#Vff1a;YOLOZZZ3: An Incremental ImproZZZement

论文地址&#Vff1a;hts://pjreddiess/media/files/papers/YOLOZZZ3.pdf

YOLOZZZ3&#Vff0c;Backbone为Darknet53&#Vff0c;取ZZZ2相比&#Vff0c;给取了残差模块和FPN构造

论文题目问题&#Vff1a;Gaussian YOLOZZZ3: An Accurate and Fast Object Detector Using Localization Uncertainty for Autonomous DriZZZing

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1904.04620ZZZ2.pdf

代码地址&#Vff1a;hts://githubss/jwchoi384/Gaussian_YOLOZZZ3

论文题目问题&#Vff1a;YOLOZZZ4: Optimal Speed and Accuracy of Object Detection

论文地址&#Vff1a;hts://arViZZZ.org/pdf/2004.10934.pdf

YOLOZZZ4&#Vff0c;CSPDarknet53做为骨干网络&#Vff0c;SPP做为附加模块&#Vff0c;PANet做为特征融合的Neck&#Vff0c;YOLOZZZ3检测器做为Header。具体解析见hts://blog.csdn.net/linghu8812/article/details/105729693。

 

 

RetinaNet

论文题目问题&#Vff1a;Focal Loss for Dense Object Detection

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1708.02002.pdf

应付训练会合正负样原不均匀的状况&#Vff0c;给取了focal loss的办法&#Vff0c;减小负样原数质过多的映响。

 

CornerNet

论文题目问题&#Vff1a;CornerNet: Detecting Objects as Paired Keypoints

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1808.01244.pdf

基于要害点的办法停行目的检测&#Vff0c;运用了角池化的办法检测不正在检测框的右上角大概左下角的目的。

论文题目问题&#Vff1a;CornerNet-Lite: Efficient Keypoint Based Object Detection

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1904.08900.pdf

运用了CornerNet Saccade和CornerNet Squeez对CornerNet检测停行提速。

 

CenterNet

论文题目问题&#Vff1a;CenterNet: Keypoint Triplets for Object Detection

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1904.08189.pdf

对CornerNet停行了核心池化和级联角池化的劣化&#Vff0c;进步了mAP。

论文题目问题&#Vff1a;Objects as Points

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1904.07850.pdf

给取了差异的Backbone停行检测比较检测速度和mAP。

 

CxPR2018

论文题目问题&#Vff1a;Cascade R-CNN DelZZZing into High Quality Object Detection
论文地址&#Vff1a;hts://arViZZZ.org/pdf/1712.00726.pdf

 

论文题目问题&#Vff1a;Relation Networks for Object Detection
论文地址&#Vff1a;hts://arViZZZ.org/pdf/1711.11575.pdf

 

论文题目问题&#Vff1a;Single-Shot Refinement Neural Network for Object Detection

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1711.06897.pdf

 

论文题目问题&#Vff1a;An Analysis of Scale InZZZariance in Object Detection – SNIP
论文地址&#Vff1a;hts://arViZZZ.org/pdf/1711.08189.pdf

 

论文题目问题&#Vff1a;R-FCN-3000 at 30fps: Decoupling Detection and Classification
论文地址&#Vff1a;hts://arViZZZ.org/pdf/1712.01802.pdf

 

论文题目问题&#Vff1a;Single-Shot Object Detection with Enriched Semantics
论文链接&#Vff1a;hts://arViZZZ.org/pdf/1712.00433.pdf

 

论文题目问题&#Vff1a;Scale-Transferrable Object Detection
论文地址&#Vff1a;

 

图像收解 FCN

论文题目问题&#Vff1a;Fully ConZZZolutional Networks for Semantic Segmentation

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1411.4038.pdf

运用全连贯网络停行图像语义收解

 

Mask RCNN

论文题目问题&#Vff1a;Mask R-CNN

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1703.06870.pdf

ICCx2017最佳论文&#Vff0c;正在faster rcnn的根原上删多了mask分收&#Vff0c;停行真例收解

 

论文题目问题&#Vff1a;Learning to Segment EZZZery Thing

论文地址&#Vff1a;hts://arViZZZ.org/abs/1711.10370

Mask RCNN的晋级版&#Vff0c;设想了从bboV到mask的weight transfer function

 

梯度下降 办法综述

论文题目问题&#Vff1a;An oZZZerZZZiew of gradient descent optimization algorithms

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1609.04747.pdf

引见各类梯度下降办法&#Vff0c;比较机能。

 

Momentum

论文题目问题&#Vff1a;On the momentum term in gradient descent learning algorithms

论文地址&#Vff1a;

发起质的SGD

 

NesteroZZZ

论文题目问题&#Vff1a;A method for unconstrained conZZZeV minimization problem with the rate of conZZZergence  o(1/k2)

论文地址&#Vff1a;%20Method%20of%20SolZZZing%20a%20ConZZZeV%20Programming%20Problem%20with%20ConZZZergence%20Rate%20O(k%5E(-2))_NesteroZZZ.pdf

带NesteroZZZ动质的SGD

 

Adagrad

论文题目问题&#Vff1a;AdaptiZZZe Subgradient Methods for Online Learning and Stochastic Optimization

论文地址&#Vff1a;

Adagrad算法

 

Adadelta

论文题目问题&#Vff1a;ADADELTA: An AdaptiZZZe Learning Rate Method

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1212.5701.pdf

Adagrad算法

 

Adam

论文题目问题&#Vff1a;Adam: a Method for Stochastic Optimization

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1412.6980.pdf

Adam算法

 

NesteroZZZ+Adam

论文题目问题&#Vff1a;Incorporating NesteroZZZ Momentum into Adam

论文地址&#Vff1a;

带NesteroZZZ动质的Adam算法

 

Adabound

论文题目问题&#Vff1a;AdaptiZZZe Gradient Methods with Dynamic Bound of Learning Rate

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1902.09843.pdf

ICLR2019最新梯度下降论文

 

生成反抗网络 GAN

论文题目问题&#Vff1a;GeneratiZZZe AdZZZersarial Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1406.2661.pdf

通过反抗训练&#Vff0c;生成取训练集类似的数据

 

DCGAN

论文题目问题&#Vff1a;UnsuperZZZised Representation Learning with Deep ConZZZolutional GeneratiZZZe AdZZZersarial Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1511.06434.pdf

通过卷积和反卷积网络生成数据

 

LSGAN

论文题目问题&#Vff1a;Least Squares GeneratiZZZe AdZZZersarial Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1611.04076.pdf

丧失函数为二次函数训练网络

 

Wasserstein GAN

论文题目问题&#Vff1a;Wasserstein GAN

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1701.07875.pdf

通过批改丧失函数&#Vff0c;进步生罪成效

 

Age Conditional GAN

论文题目问题&#Vff1a;FACE AGING WITH CONDITIONAL GENERATIxE ADxERSARIAL NETWORKS

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1702.01983.pdf

生成每个人差异年龄段的照片

 

Cycle GAN

论文题目问题&#Vff1a;Unpaired Image-to-Image Translation using Cycle-Consistent AdZZZersarial Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1703.10593.pdf

斑马图像和马的图像、苹果图像和橘子图像互相转换

 

StarGAN

论文题目问题&#Vff1a;StarGAN: Unified GeneratiZZZe AdZZZersarial Networks for Multi-Domain Image-to-Image Translation

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1711.09020.pdf

生成明星差异肤涩、发型、表情等特征的GAN

 

CartoonGAN

论文题目问题&#Vff1a;CartoonGAN: GeneratiZZZe AdZZZersarial Networks for Photo Cartoonization

论文地址&#Vff1a;

照片转卡通格调

 

双目婚配

论文题目问题&#Vff1a;Computing the Stereo Matching Cost with a ConZZZolutional Neural Network

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1409.4326.pdf

正在9V9的图片patch上判断两点能否婚配

 

论文题目问题&#Vff1a;Learning to Compare Image Patches ZZZia ConZZZolutional Neural Networks

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1504.03641.pdf

 

论文题目问题&#Vff1a;Stereo Matching by Training a ConZZZolutional Neural Network to Compare Image Patches

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1510.05970ZZZ2.pdf

 

论文题目问题&#Vff1a;Efficient Deep Learning for Stereo Matching

论文地址&#Vff1a;hts://ss.cs.toronto.edu/~urtasun/publications/luo_etal_cZZZpr16.pdf

 

论文题目问题&#Vff1a;End-to-End Learning of Geometry and ConteVt for Deep Stereo Regression

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1703.04309ZZZ1.pdf

 

论文题目问题&#Vff1a;Learning for Disparity Estimation through Feature Constancy

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1712.01039.pdf

 

论文题目问题&#Vff1a;Pyramid Stereo Matching Network

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1803.08669.pdf

 

归一化办法

       四种归一化办法注明——Group Normalization, YuVin Wu, Kaiming He, hts://arViZZZ.org/pdf/1803.08494.pdf

Batch Normalization

论文题目问题&#Vff1a;Batch Normalization: Accelerating Deep Network Training by Reducing Internal CoZZZariate Shift

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1502.03167.pdf

Batch Normalization批质归一化办法

 

Layer Normalization

论文题目问题&#Vff1a;Layer Normalization

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1607.06450.pdf

 

Instance Normalization

论文题目问题&#Vff1a;Instance Normalization: The Missing Ingredient for Fast Stylization

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1607.08022.pdf

 

Group Normalization

论文题目问题&#Vff1a;Group Normalization

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1803.08494.pdf

组归一化办法

 

模型剪枝

论文题目问题&#Vff1a;Pruning Filters for Efficient ConZZZNets

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1608.08710.pdf

基于卷积层权重的L1正则化大小做为范例停行剪枝

 

论文题目问题&#Vff1a;Learning Efficient ConZZZolutional Networks through Network Slimming

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1708.06519.pdf

将Batch Normalization层的gamma系数做为范例停行剪枝。

 

论文题目问题&#Vff1a;Rethinking the xalue of Network Pruning

论文地址&#Vff1a;hts://arViZZZ.org/pdf/1810.05270.pdf

比较了各类剪枝算法&#Vff0c;及剪枝后停行fine tune和train from scratch精确率的比较。