图像语义分割

https://handong1587.github.io/deep_learning/2015/10/09/segmentation.html

http://blog.csdn.net/column/details/15893.html

pascal_voc语义分割排行榜http://host.robots.ox.ac.uk:8080/leaderboard/displaylb.php?challengeid=11&compid=6

图像分类，物体检测，语义分割，实例分割的联系和区别

图像分割的性能测试指标小结

从特斯拉到计算机视觉之「图像语义分割」

关于图像语义分割的总结和感悟

反卷积与语义分割

全卷积网络（FCN）与图像分割

语义分割相关总结

FCN

Fully Convolutional Networks for Semantic Segmentation

深度学习入门:Fully Convolutional Networks

深度学习入门:Fully Convolutional Networks for Semantic Segmentation

Fully Convolutional Networks for Semantic Segmentation

深度学习（二十一）基于FCN的图像语义分割-CVPR 2015

全卷积网络 FCN 详解

1. FCN for Sematic Segmentation.pptx

全卷积网络：从图像级理解到像素级理解

Semantic Segmentation with FCN里的32x, 16x和8x

语义分割--FCN 算法中的一些细节--特征怎么融合

deeplab

项目主页http://liangchiehchen.com/projects/DeepLab.html

Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs

论文阅读笔记：图像分割方法deeplab以及Hole算法解析

DeepLab论文解读

深度学习之DeepLab用于语义分割

[Paper Reading] DeepLab v1 & v2

几分钟看完语义分割经典论文:DeepLab

DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs

CRF图像分割简介

https://hsh.blog.ustc.edu.cn/wp-content/uploads/2015/09/Class-Segmentation-and-Object-Localization-with-Superpixel-Neighborhoods.pdf

推荐一些用CRF做图像语义分割的资源

Conditional Random Fields as Recurrent Neural Networks

深度学习（三十三）CRF as RNN语义分割-未完待续

CRF as RNN的原理及Caffe实现

图像语义分割之FCN和CRF

CRF as RNN 代码解读

1. Semantic Image Segmentation With Deep Convolutional Nets and Fully Connected CRFs.ppt

http://www.robots.ox.ac.uk/~szheng/CRFasRNN.html

U-Net

全卷机神经网络图像分割-keras实现

DeconvNet

Learning Deconvolution Network for Semantic Segmentation

MNC

《Instance-aware Semantic Segmentation via Multi-task Network Cascades》(MNC)，用自己的数据运行

ＭＮＣ用自己的数据运行得到如下错误：IndexError: index 4 is out of bounds for axis 1 with size 4

论文解读-

图像分割“Instance-aware Semantic Segmentation via Multi-task Network Cascades”

ParseNet

ParseNet论文解读

语义分割(semantic segmentation) 常用神经网络介绍对比-FCN SegNet U-net DeconvNet

多篇用DL做Semantic Segmentation的文章总结

PSPNet

深度学习之PSPnet用于语义分割

几分钟看完PSPnet：Pyramid Scene Parsing Network

Pyramid Scene Parsing Network - arxiv - 1612.01105

图像语义分割之特征整合和结构预测

ICNet

实时语义分割--ICNet for Real-Time Semantic Segmentation on High-Resolution Images

https://github.com/hszhao/ICNet

http://blog.csdn.net/yaoliangsjtu/article/details/77620936

弱标注

Weakly- and Semi-Supervised Learning of a Deep Convolutional Network for Semantic Image Segmentation

CNN在基于弱监督学习的图像分割中的应用

论文笔记 | BoxSup: Exploiting Bounding Boxes to Supervise Convolutional Networks for Semantic Segmentati

An Adversarial Regularisation for Semi-Supervised Training of Structured Output Neural Networks

弱监督学习在语义分割中的进展与应用?

和GAN结合

FCNs in the Wild: Pixel-level Adversarial and Constraint-based Adaptation

Semantic Segmentation using Adversarial Networks

深度对抗学习在图像分割和超分辨率中的应用

论文精读 Learning to Refine Object Segments

基于Mutil-Scale CNN的图片语义分割、法向量估计-ICCV 2015

Simultaneous Detection and Segmentation (SDS)

Hypercolumn Object Segmentation

[Paper Reading] Attention to Scale: Scale-aware Semantic Image Segmentation

几分钟看完 Zifeng Wu的38层网络：Wider or Deeper: Revisiting the ResNet Model for Visual Recognition

几分钟看完RefineNet:RefineNet: Multi-Path Refinement Networks for High-Resolution Semantic Segmentation

几分钟看完zifeng wu instance level segmentation：Bridging Category-level and Instance-level Semantic Image

几分钟看完旷视孙剑最新大作-Large Kernel Matters —— Improve Semantic Segmentation by Global Convolutional Network

论文笔记：Label Refinement Network for Coarse-to-Fine Semantic Segmentation

Fully Convolutional Instance-aware Semantic Segmentation 论文笔记

将CRF的refine替换为边缘检测方法

Semantic Image Segmentation with Task-Specific Edge Detection Using CNNs and a Discriminatively Trained Domain Transform

http://blog.csdn.net/zhangjunhit/article/category/6809642

语义分割 - Semantic Segmentation Papers

FCN实践

Ubuntu下全卷积神经网络（FCN）框架的配置和基本使用

FCN网络训练 终极版

使用FCN做图像语义分割(实践篇)

net_surgery中如何将全连接层转换成卷积层

在caffe上做FCN模型fine-tune的一些注意事项

FCN网络的训练——以SIFT-Flow 数据集为例

FCN模型训练中遇到的困难

FCN网络训练 SIFTFLOW数据集

Tensoflow学习记录13--用深度学习来做图像分割Fully Convolutional Networks for Semantic Segmentation (FCNs)

FCN中的transplant

以fcn.berkeleyvision.org的sift-flow-fcn32s为例，训练自己的数据集时除了上面说的要改的地方，还有几个需要注意

1.图像的通道数，如果在VGG上面finetuning的话要求输入通道为3

2.不需要label_geometric，所以要把与它相关的地方都注释掉

（1）siftflow_layers.py中搜索label_geometric，注释掉，if len(top) != 3:改为if len(top) != 2:

（2）trainval.prototxt与test.prototxt中top: "geo"这行注释掉，这个不注释掉会出现check failed:data_的错误

layer {
  name: "loss"
  type: "SoftmaxWithLoss"
  bottom: "score_sem"
  bottom: "sem"
  top: "loss"
  loss_param {
    ignore_label: 255
    normalize: false
  }
}这层以后的和geo相关的层全部注释掉

（3）solve.py中score.seg_tests(solver, False, test, layer='score_geo', gt='geo')这句注释掉

3.几个prototxt文件中最后输出的feature map的num_ouput改成类别数+1

4.改数据输入路径，siftflow_layers.py中的load_image和load_label函数加载图像和label的路径改一下，对应prototxt的param_str: "{\'siftflow_dir\':改成自己的数据路径，solve.py中test = np.loadtxt这句改成test.txt的路径

像素级别的loss的取均值决定learning_rate的数量级. 使用224×224的图片, 如果loss是对单个像素的均值, 则数量级在10−5. 如果是对单张图片的loss, 要再小4个数量级.(caffe fcn使用了10−10)，这里是fcn的solver.prototxt。

deeplab v2实践

DeepLab V2 配置bug

DeepLab v2 配置

DeepLab V2安装配置

crfasrnn环境搭建，跑demo

资源

https://github.com/shelhamer/fcn.berkeleyvision.org

https://github.com/torrvision/crfasrnn

https://github.com/martinkersner/train-CRF-RNN

https://github.com/wkentaro/fcn

https://github.com/ycszen/pytorch-ss

https://github.com/msracver/FCIS