目標檢測之CNN系列
排行榜
pascal VOC:http://host.robots.ox.ac.uk:8080/leaderboard/main_bootstrap.php
coco:http://mscoco.org/dataset/#detections-leaderboard
kitti:http://www.cvlibs.net/datasets/kitti/eval_object.php
http://blog.csdn.net/maweifei/article/details/59078077
http://blog.csdn.net/maweifei/article/details/59078077
overfeat
RCNN
DeepID-Net
selective search & Edge box
其實·RPN也算一種
sppnet
Fast RCNN
Faster RCNN
http://wenku.baidu.com/link?url=11zlcxN9p7d6ydhJHnDCBltkS9HEyLLZ0sBBgdwq7Oa02BsXeKRIBvDWcPVzKMFV8SOBtc9qdhTilohJ55MhUc7Ht7jDUiCh4yhn5xvRUYMiXX2T9lzG2zdYF5PDuFtn
http://blog.csdn.net/qq_26898461/article/category/6204814
DeepBox
YOLO
http://blog.csdn.net/u012235274/article/category/6204386
yolo v2 訓練
DenseBox
SSD
http://blog.csdn.net/u012235274/article/category/6366885
G-CNN
MultiPathNet
https://github.com/facebookresearch/multipathnet
HyperNet
LocNet
OHEM
https://github.com/abhi2610/ohem
R-FCN
https://github.com/Orpine/py-R-FCN
MS-CNN
https://github.com/zhaoweicai/mscnn
PVANET
FPN
RRC
focal loss
focal loss
http://blog.csdn.net/zhangjunhit/article/category/6647655
Is Faster R-CNN Doing Well for Pedestrian Detection?
論文 Is Faster R-CNN Doing Well for Pedestrian Detection?探討在行人檢測領域Faster R-CNN是否有效，提出了RPN + Boosted Forest分類器。
基於R-CNN的多尺度改進方法概述
梳理基於R-CNN的多尺度改進方法，主要思路是提取多個層的feature進行卷積層的特徵融合(即skip connections),涉及的方法有MultiPath Network，ION(Inside-Outside Net)，HyperNet，PVANET及MS-CNN。
多尺度R-CNN論文筆記(1): A MultiPath Network for Object Detection
多尺度R-CNN論文筆記(2): Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks
多尺度R-CNN論文筆記(3): HyperNet: Towards Accurate Region Proposal Generation and Joint Object Detection
多尺度R-CNN論文筆記(4): PVANET: Deep but Lightweight Neural Networks for Real-time Object Detection
多尺度R-CNN論文筆記(5): A Unified Multi-scale Deep Convolutional Neural Network for Fast Object Detection
多尺度R-CNN論文筆記(6): Feature Pyramid Networks for Object Detection
無proposal，基於迴歸的檢測演算法概述
開創性工作YOLO與後續改進，以及G-CNN，SSD等工作。
無proposal檢測方法(1): You Only Look Once: Unified, Real-Time Object Detection
無proposal檢測方法(2): G-CNN: an Iterative Grid Based Object Detector
無proposal檢測方法(3): SSD: Single Shot MultiBox Detector
YOLO9000: Better,Faster,Stronger（YOLO9000：更好，更快，更強）
和GAN結合
弱監督/無監督
Weakly Supervised Deep Detection Networks
Weakly supervised localization of novel objects using appearance transfer
ContextLocNet: Context-Aware Deep Network Models for Weakly Supervised Localization
Weakly Supervised Cascaded Convolutional Networks
https://github.com/jbhuang0604/WSL
VOC2007中包含9963張標註過的圖片， 由train/val/test三部分組成， 共標註出24,640個物體。
VOC2007的test資料label已經公佈， 之後的沒有公佈（只有圖片，沒有label）。
對於檢測任務，VOC2012的trainval/test包含08-11年的所有對應圖片。 trainval有11540張圖片共27450個物體。
對於分割任務， VOC2012的trainval包含07-11年的所有對應圖片， test只包含08-11。trainval有 2913張圖片共6929個物體。
spp_net實踐
RCNN實踐
淺析py-faster-rcnn中不同版本caffe的安裝及其對應不同版本cudnn的解決方案Faster rcnn 安裝、訓練、測試、除錯
faster rcnn demo.py:在一個視窗顯示所有類別標註
http://blog.csdn.net/10km/article/category/6816967
faster rcnn multi GPU
https://www.google.com/search?sclient=psy-ab&site=&source=hp&btnG=Search&q=faster+rcnn+multi+gpu
https://github.com/rbgirshick/py-faster-rcnn/issues/143
https://github.com/bharatsingh430/py-R-FCN-multiGPU
https://github.com/endernewton/tf-faster-rcnn
https://github.com/msracver/Deformable-ConvNets
https://github.com/bharatsingh430/Deformable-ConvNets/
yolo實踐
https://github.com/xingwangsfu/caffe-yolo
https://pjreddie.com/darknet/yolo/
http://guanghan.info/blog/en/my-works/train-yolo/
SSD實踐
https://github.com/balancap/SSD-Tensorflow