深度学习从入门到放弃之CV-Semantic Segmentation目录

目录

一、常用数据集

二、典型思路

三、经典论文目录


一、常用数据集

PASCAL VOC 2012 1.5k训练图像,1.5k验证图像,20个类别(包含背景)。

MS COCO COCO比VOC更困难。有83k训练图像,41k验证图像,80k测试图像,80个类别。

二、典型思路


Semantic Segmentation目前已经被深度学习占领,各种模型层出不穷。个人目前主要精读了一些2D Semantic image Segmentation,在技术实现上大体分为3派:

1、FCN(Fully Convolutional Networks),通过Encoder-Decoder模型建立end-to-end的训练,技术关键点deconvolution/Transposed convolution(后者描述更确切)。

2、CRF(Conditional random fields)/MRF(Markov random fields)

3、Dilated Convolutions,带孔的卷积

本人会挑拣一些经典论文写一下阅读笔记。


state-of-the-art更新中:

1、Segmentation Results: VOC2012:

三、经典论文目录

个人总结的经典论文目录:

2015----FCN----Fully Convolutional Networks for Semantic Segmentation

2015----Learning Deconvolution Network for Semantic Segmentation

2015----U-Net_Convolutional Networks for Biomedical

2016----Deeplab V1----Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs

2016----DeepLab V2----DeepLab Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs

2016----ENet_A Deep Neural Network Architecture for Real-Time Semantic Segmentation

2016----Laplacian Pyramid Reconstruction and Refinement for Semantic Segmentation

2016----Learning to Refine Object Segments

2016----Multi-Scale Context Aggregation by Dilated Convolutions

2016----RefineNet_Multi-Path Refinement Networks for High-Resolution Semantic Segmentation

2016----SegNet_A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation

2017---- The One Hundred Layers Tiramisu_Fully Convolutional DenseNets for Semantic Segmentation

2017----Deeplab V3----Rethinking Atrous Convolution for Semantic Image Segmentation

2017----ICNet for Real-Time Semantic Segmentation on High-Resolution Images

2017----Large Kernel Matters -- Improve Semantic Segmentation by Global Convolutional Network

2017----Loss Max-Pooling for Semantic Image Segmentation

2017----Pyramid Scene Parsing Network

2017----Understanding Convolution for Semantic Segmentation

2017----SDN----Stacked Deconvolutional Network for Semantic Segmentation

2018----Deeplabv3+----Encoder-Decoder with Atrous Separable Convolution for Semantic Image Segmentation

2018----Searching for Efficient Multi-Scale Architectures for Dense Image Prediction arXiv:1809.04184 (2018)

Google 在 Cloud AutoML 不断发力,相比较而言之前的工作只是在图像分类领域精耕细作,如今在图像分割
开疆扩土,在 arxiv 提交第一篇基于 NAS(Neural network architecture)的语义分割模型(DPC,dense 
prediction cell)已经被 NIPS2018 接收,并且在 Cityscapes,PASCAL-Person-Part,PASCAL VOC 2012 
取得 state-of-art 的性能(mIOU 超过 DeepLabv3+)和更高的计算效率(模型参数少,计算量减少)。

返回CV总目录

编辑于 2018-10-09

文章被以下专栏收录