CNNR-CNN
A Brief History of CNNs in Image Segmentation: From R-CNN to Mask R-CNN
FCNFully Convolutional Network
|
AthelasConvolutional Neural NetworksCNNs
CNNs
Alex Krizhevsky, Geoff Hinton, and Ilya SutskeverImageNet 2012CNNs
CNNsImageNet
1 ImageNetCNNsYImageNet
2
2
3
4
[1703.06870] Mask R-CNN
CNNs
CNNsRoss Girshick
R-
CNNRegional CNNCNNsFast R-CNNFaster R-
CNNFacebook ResearchMask R-CNN
1. R-CNN: [1311.2524] Rich feature hierarchies for accurate object detection and
semantic segmentation
2. Fast R-CNN: [1504.08083] Fast R-CNN
3. Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
4. Mask R-CNN: [1703.06870] Mask R-CNN
2014: R-CNN - An Early Application of CNNs to Object Detection
5 R-CNN
TorontoHintonUC BerkeleyJitendra Malik
Krizhevsky
5Ross Girshick (a name we’ll
see again)Jeff DonahueTrevor DarrelKrizhevsky
PASCAL VOCImageNet
PASCAL VOCCNNs
HOG
Regions CNNsR-CNN
R-CNN
R-CNNbounding box
Bounding Box+
bounding boxR-CNNbox
box
6 Selective Search
https://www.koen.me/research/pub/uijlings-ijcv2013-draft.pdf
R-CNNSelective Searchregion proposalsSelective Search
6
7 region proposalsR-CNNAlexNetregion proposals
region
[1311.2524] Rich feature hierarchies for accurate object detection and semantic segmentation
region proposalsR-CNNAlexNet
region7
CNNR-CNNSupport Vector MachineSVM
Bounding Box
Boxtighten bounding box
R-CNNR-CNNregion
proposalstightenBounding Box
sub-regions
sub-resionsBounding Box
R-CNN
1. regions proposalsBounding Box
2. AlexNet+SVMBouding Box
3. Bounding BoxBoxtighter bounding boxes
2015: Fast R-CNN - Speeding up and Simplifying R-CNN
R-CNN
1. region proposalCNNAlexNet
2000CNN
2. ——CNNtighten
bounding box
2015R-CNNRoss GirshickFast R-CNN
Fast R-CNN
RoIRegion of InterestPooling
Ross Girshickproposed regionsCNN
2000——CNN2000proposed
regions
8 RoIPoolCNNregion
Stanford’s CS231N slides by Fei Fei Li, Andrei Karpathy, and Justin
Johnson
Fast R-CNNRoIPoolRegion of Interest PoolingRoIPool
CNNregions8CNNCNN
poolmax
poolingCNNR-CNN2000
Combine All Models into One Network
9 Fast R-CNNCNN
https://www.slideshare.net/simplyinsimple/detection-52781995
Fast R-CNNCNNR-CNN
CNNSVMtightenFast R-
CNN
9Fast R-CNNsoftmax layerSVMCNN
softmax layerlinear regression layer
region proposals
tighter bounding boxes
2016: Faster R-CNN - Speeding Up Region Proposal
Fast R-CNN——region proposer
bouding boxregions of interestFast R-CNNSelective
Search
2015Shaoqing Ren, Kaiming He, Ross Girshick, and Jian Sun
Faster R-CNNregion proposal
Faster R-CNNregion proposalsCNN
CNNregion proposals
selective search
10 Faster R-CNNCNNregion proposalsTowards
Real-Time Object Detection with Region Proposal Networks
Faster R-CNN10CNNregion proposals
CNNregion proposals
Fast R-CNNregion
proposalsregion proposals
region proposals
bounding box
How the Regions are Generated
Faster R-CNNCNNregion proposalsFaster R-CNNCNN
Region Proposal Network.