目标检测进化史

one-stage:

再之前的就看这个啦:

Dave:基于深度学习的「目标检测」算法综述zhuanlan.zhihu.com图标


RefineDet:

vision:RefineDet 论文解析zhuanlan.zhihu.com图标ddwang:RefineDet网络解析以及PyTorch实现zhuanlan.zhihu.com图标


RetinaNet:

TeddyZhang:目标检测:RetinaNet(ICCV 2017)zhuanlan.zhihu.com图标manofmountain:RetinaNet: Focal loss在目标检测中的应用zhuanlan.zhihu.com图标究竟灰:读Focal Losszhuanlan.zhihu.com图标

Repulsion loss:

极市平台:CVPR18|Repulsion loss:遮挡下的行人检测zhuanlan.zhihu.com图标


RFBNet:

胡孟:RFBNet(1)_论文_ECCV2018zhuanlan.zhihu.com图标

KL-Loss:

飘哥:学习CVPR 2019 论文《用于目标检测的不确定性边界框》zhuanlan.zhihu.com图标

CornerNet:飘哥:学习CVPR 2019 论文《用于目标检测的不确定性边界框》CornerNet:

王若霄:CornerNet算法解读zhuanlan.zhihu.com图标


ScratchDet:

张凯:2019 CVPR目标检测论文ScratchDet(设计检测的backbone)zhuanlan.zhihu.com图标


三分支网络:

EdisonGzq:三分支网络——目前目标检测性能最佳网络框架zhuanlan.zhihu.com图标

目标检测 FSAF:

ChenJoya:CVPR2019 | 目标检测 FSAF:为金字塔网络的每一层带去最好的样本zhuanlan.zhihu.com图标ywsun:[CVPR2019]:FSAF for Single-Shot Object Detectionzhuanlan.zhihu.com图标

ScratchDet:

机器之心:CVPR 2019 | 京东AI研究院提出 ScratchDet:随机初始化训练SSD目标检测器zhuanlan.zhihu.com图标

SimpleDet:

Naiyan Wang:SimpleDet: 一套简单通用的目标检测与物体识别框架zhuanlan.zhihu.com图标

TridentNet:

Naiyan Wang:TridentNet:处理目标检测中尺度变化新思路zhuanlan.zhihu.com图标路一直都在:【CVPR2019论文阅读】:TridentNetzhuanlan.zhihu.com图标

M2Det:

张凯:2019AAAI目标检测论文M2Det(One-stage算法)zhuanlan.zhihu.com图标joking:简单教教M2Detzhuanlan.zhihu.com图标

ExtremeNet:

Amusi:德克萨斯大学提出:One-stage目标检测最强算法 ExtremeNetzhuanlan.zhihu.com图标


GIoU:

刘浪:斯坦福CVPR2019 GIoU:目标检测任务的新Losszhuanlan.zhihu.com图标

基于区域分解&集成的目标检测:

EdisonGzq:性能大幅度提升(速度&遮挡) | 基于区域分解&集成的目标检测zhuanlan.zhihu.com图标https://arxiv.org/pdf/1901.08225.pdfarxiv.org

旷视科技提出:最新实时目标检测网络 ThunderNet

旷视科技提出:最新实时目标检测网络 ThunderNetbbs.cvmart.net图标

商汤科技:目标检测中的特征交织机制:

yuanCruise:商汤科技:目标检测中的特征交织机制zhuanlan.zhihu.com图标

FCOS: 最新的one-stage逐像素目标检测算法:

yuanCruise:FCOS: 最新的one-stage逐像素目标检测算法zhuanlan.zhihu.com图标Amusi:最新的Anchor-Free目标检测模型FCOS,现已开源!zhuanlan.zhihu.com图标

Libra R-CNN(一个全面平衡的目标检测器):

如何看待 CVPR2019 论文 Libra R-CNN(一个全面平衡的目标检测器)?www.zhihu.com图标https://mp.weixin.qq.com/s?__biz=MzUxNjcxMjQxNg==&mid=2247488721&idx=2&sn=bf3be97c413311281d4d9b97b755a847&chksm=f9a2665eced5ef48e748b2de557b585faf63895c61b68fc47169903f634d866c11dea6ce7f13&mpshare=1&scene=1&srcid=&key=eabd007c9b31d563b824fdec2abce33b50b9b94016fa85046265886aa944bc2b87140573665f0307d83b3d2891f2eea0eac2e204539499530aa7406179fbb23300baa927d089ea841de952d30542a437&ascene=1&uin=MjAyODc3ODcyMg%3D%3D&devicetype=Windows+10&version=62060739&lang=zh_CN&pass_ticket=nFmowjXjS9sSsMx8R%2Fhu7UN0jiDyhaisxhwH%2BbO1Exc0eP8OPDcwU7ngtE7703pgmp.weixin.qq.com


https://arxiv.org/pdf/1904.02701.pdfarxiv.org
张凯:2019CVPR Libra RCNN目标检测算法(特征融合)zhuanlan.zhihu.com图标kynov:论文阅读(改进不平衡问题)zhuanlan.zhihu.com图标


anchor-free目标检测模型FoveaBox:

如何评价最新的anchor-free目标检测模型FoveaBox?www.zhihu.com图标https://arxiv.org/pdf/1904.01355.pdfarxiv.org

CenterNet:将目标视为点 (已开源):

https://zhuanlan.zhihu.com/p/62724053zhuanlan.zhihu.com图标Amusi:CenterNet:将目标视为点 (已开源)zhuanlan.zhihu.com图标kwduan:中科院牛津华为诺亚提出CenterNet,one-stage detector可达47AP,已开源!zhuanlan.zhihu.com图标autocyz:基于点检测的物体检测方法(四):CenterNetzhuanlan.zhihu.com图标OLDPAN:扔掉anchor!真正的CenterNet——Objects as Points论文解读zhuanlan.zhihu.com图标

CornerNet-Lite:

Amusi:超越YOLOv3!普林斯顿大学提出:CornerNet-Lite,基于关键点的目标检测算法,已开源!zhuanlan.zhihu.com图标


何恺明最新论文:VoteNet 3D目标检测:

https://arxiv.org/abs/1904.09664arxiv.org
Deep Hough Voting for 3D Object Detection in Point Cloudsarxiv.org
https://mp.weixin.qq.com/s?__biz=MzUxNjcxMjQxNg==&mid=2247488790&idx=1&sn=c8076273ae15d20c6fed5c72d3885782&chksm=f9a26799ced5ee8fe40d59d286b44481a24b6f2a4d1334756784ad356a992da3860d51aec94f&mpshare=1&scene=1&srcid=&key=593393174013ce6dcfd0c23299b7e3e3e36a13e184ff471b6b9f2e7b8e64c505f4218481ebbb6bdf2a0ad43c60d590e51b051de4120fd2a4501102668b4b4f04a8aa3aa4f215f40970d90fd761321a07&ascene=1&uin=MjAyODc3ODcyMg%3D%3D&devicetype=Windows+10&version=62060739&lang=zh_CN&pass_ticket=CE6UNmztFX0BsUPG2XuQS%2FTgPlTkLbbf0tJCK%2FMXrjMbz8Um1DnSDgFTuHWjWurcmp.weixin.qq.com
新智元:何恺明团队最新研究:3D目标检测新框架VoteNet,直接处理点云数据,刷新最高精度zhuanlan.zhihu.com图标

PoolNet:

A Simple Pooling-Based Design for Real-Time Salient Object Detectionarxiv.org
A Simple Pooling-Based Design for Real-Time Salient Object Detectionmmcheng.net

RepPoints:

https://mp.weixin.qq.com/s?__biz=MzUxNjcxMjQxNg==&mid=2247488945&idx=2&sn=94cd193000d1918c743541a3035bcd47&chksm=f9a2673eced5ee2892a399007e1a9a4ff491e3ef33a4d46b63da8aef2f1cf26c60b85c005367&mpshare=1&scene=1&srcid=&key=b2fabf0b695501ec09d7c599652299be6689149641f1aebd1829a73caf773305ac5a80d370e00e8e6f177f4d671a437d0e20cc277b2647a7f955abcb9aabfdcc7075c7e43dc40997a44176e22756cde8&ascene=1&uin=MjAyODc3ODcyMg%3D%3D&devicetype=Windows+10&version=62060739&lang=zh_CN&pass_ticket=O5cPeOq8tSS52vZvpG44IIHk%2BnXkxvLRdpi6Xps92h0VN4Y%2FSZVoJPd4yv36SwrRmp.weixin.qq.com
https://arxiv.org/pdf/1904.11490.pdfarxiv.org
陀飞轮:目标检测:可形变卷积的进阶zhuanlan.zhihu.com图标Tenacious:RepPoints: Point Set Representation for Detectionzhuanlan.zhihu.com图标

Light-Weight RetinaNet:

https://arxiv.org/pdf/1905.10011.pdfarxiv.org

An Analysis of Scale Invariance in Object Detection - SNIP:

麦田守望者:SNIP解读zhuanlan.zhihu.com图标

Anchor free深度学习的目标检测方法:

黄浴:Anchor free深度学习的目标检测方法zhuanlan.zhihu.com图标

从Densebox到Dubox:更快、性能更优、更易部署的anchor-free目标检测:

Amusi:从Densebox到Dubox:更快、性能更优、更易部署的anchor-free目标检测zhuanlan.zhihu.com图标

FoveaBox: Beyond Anchor-based Object Detector:

路一直都在:FoveaBox: Beyond Anchor-based Object Detectorzhuanlan.zhihu.com图标

解读目标检测新范式:Segmentations is All You Need:

机器之心:解读目标检测新范式:Segmentations is All You Needzhuanlan.zhihu.com图标

YOLOv3+: A Learning Technique to Improve Object

chaser:YOLOv3+: A Learning Technique to Improve Objectzhuanlan.zhihu.com图标

Mask Scoring R-CNN:

ManWingloeng:CVPR2019 | Mask Scoring R-CNN 论文解读zhuanlan.zhihu.com图标

2019 AAAI GHM(解决one-stage样本不平衡问题):

张凯:2019 AAAI GHM(解决one-stage样本不平衡问题)目标检测算法论文阅读笔记zhuanlan.zhihu.com图标

Grid R-CNN Plus(2019):

TeddyZhang:目标检测:Grid R-CNN Plus(2019)zhuanlan.zhihu.com图标

GCNet: Non-local Networks Meet Squeeze-Excitation Networks and Beyond:

张凯:2019 GCNet(attention机制,目标检测backbone性能提升)论文阅读笔记zhuanlan.zhihu.com图标

张凯:2019 GCNet(attention机制,目标检测backbone性能提升)论文阅读笔记

张凯:2019 GCNet(attention机制,目标检测backbone性能提升)论文阅读笔记zhuanlan.zhihu.com图标

Cascade R-CNN: High Quality Object Detection and Instance Segmentation:

https://arxiv.org/pdf/1906.09756.pdfarxiv.org

2019CVPR Libra RCNN目标检测算法(特征融合):

张凯:2019CVPR Libra RCNN目标检测算法(特征融合)zhuanlan.zhihu.com图标

Double-Head RCNN:

chaser:Double-Head RCNNzhuanlan.zhihu.com图标

Data Augmentation+NAS-FPN:

如何评价谷歌大脑最新关于目标检测的Data Augmentation论文?www.zhihu.com图标

SSA-CNN:

张凯:2019 CVPR SSA-CNN(自注意力机制)目标检测算法论文阅读笔记zhuanlan.zhihu.com图标

Revisiting Feature Alignment for One-stage Object Detection:

https://arxiv.org/pdf/1908.01570.pdfarxiv.org

Matrix Nets: A New Deep Architecture for Object Detection:

https://arxiv.org/pdf/1908.04646.pdfarxiv.org
机器之心:参数少一半、速度快3倍:最新目标检测核心架构来了zhuanlan.zhihu.com图标

OD-GCN: OBJECT DETECTION BY KNOWLEDGE GRAPH WITH GCN :

https://arxiv.org/ftp/arxiv/papers/1908/1908.04385.pdfarxiv.org

AlignDet:

张凯:2019 AlignDet(One-stage目标检测算法,mAP=44.1)论文阅读笔记zhuanlan.zhihu.com图标


CBNet: A Novel Composite Backbone Network Architecture for Object Detection:

https://arxiv.org/pdf/1909.03625.pdfarxiv.org
chaser:CBNet论文解读zhuanlan.zhihu.com图标





two-stages:

。。。

Faster RCNN:

白裳:一文读懂Faster RCNNzhuanlan.zhihu.com图标

R-FCN:

麦田守望者:详解R-FCNzhuanlan.zhihu.com图标

Light-Head R-CNN:

TeddyZhang:目标检测:Light-Head R-CNN(2017)zhuanlan.zhihu.com图标

Mask RCNN:

stone:令人拍案称奇的Mask RCNNzhuanlan.zhihu.com图标


Cascade R-CNN:

过若干:Cascade R-CNN 详细解读zhuanlan.zhihu.com图标胡孟:Cascade RCNN(1)_论文_CVPR2018zhuanlan.zhihu.com图标

Training-Time-Friendly Network for Real-Time Object Detection:

https://arxiv.org/pdf/1909.00700.pdfarxiv.org

编辑于 2019-09-15

文章被以下专栏收录