【目标识别】深度学习进行目标识别的资源列表：O网页链接 包括RNN、MultiBox、SPP-Net、DeepID-Net、Fast R-CNN、DeepBox、MR-CNN、Faster R-CNN、YOLO、DenseBox、SSD、Inside-Outside Net、G-CNN等。
Papers
Deep Neural Networks for Object Detection
OverFeat: Integrated Recognition, Localization and Detection using Convolutional Networks
[td]
|
method
|
ILSVRC 2013 mAP
|
|
OverFeat
|
24.3%
|
- intro: A deep version of the sliding window method, predicts bounding box directly from each location of the topmost feature map after knowing the confidences of the underlying object categories.
- arXiv: http://arxiv.org/abs/1312.6229
- code: https://github.com/sermanet/OverFeat
R-CNN
Rich feature hierarchies for accurate object detection and semantic segmentation(R-CNN)
[td]
|
method
|
VOC 2007 mAP
|
VOC 2010 mAP
|
VOC 2012 mAP
|
ILSVRC 2013 mAP
|
|
R-CNN,AlexNet
|
54.2%
|
50.2%
|
49.6%
|
|
| R-CNN,bbox reg,AlexNet |
58.5%
|
53.7%
|
53.3%
|
31.4%
|
|
R-CNN,bbox reg,ZFNet
|
59.2%
|
|||
| R-CNN,VGG-Net |
62.2%
|
|||
| R-CNN,bbox reg,VGG-Net |
66.0%
|
- arXiv: http://arxiv.org/abs/1311.2524
- slides: http://www.image-net.org/challenges/LSVRC/2013/slides/r-cnn-ilsvrc2013-workshop.pdf
- slides: http://www.cs.berkeley.edu/~rbg/slides/rcnn-cvpr14-slides.pdf
- code: https://github.com/rbgirshick/rcnn
- notes: http://zhangliliang.com/2014/07/23/paper-note-rcnn/
- caffe-pr(“Make R-CNN the Caffe detection example”):https://github.com/BVLC/caffe/pull/482
MultiBox
Scalable Object Detection using Deep Neural Networks (MultiBox)
- intro: Train a CNN to predict Region of Interest.
- arXiv: http://arxiv.org/abs/1312.2249
- code: https://github.com/google/multibox
SPP-Net
Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition
[td]
| method |
VOC 2007 mAP
|
ILSVRC 2013 mAP
|
|
SPP_net(ZF-5),1-model
|
54.2%
|
31.84%
|
|
SPP_net(ZF-5),2-model
|
60.9%
|
|
| SPP_net(ZF-5),6-model | 35.11% |
- arXiv: http://arxiv.org/abs/1406.4729
- code: https://github.com/ShaoqingRen/SPP_net
- notes: http://zhangliliang.com/2014/09/13/paper-note-sppnet/
Learning Rich Features from RGB-D Images for Object Detection and Segmentation
Scalable, High-Quality Object Detection
DeepID-Net
DeepID-Net: Deformable Deep Convolutional Neural Networks for Object Detection
[td]
|
method
|
VOC 2007 mAP
|
ILSVRC 2013 mAP
|
|
DeepID-Net
|
64.1%
|
50.3%
|
Object Detection Networks on Convolutional Feature Maps
[td]
|
method
|
Trained on
|
mAP
|
|
NoC
|
07+12
|
68.8%
|
|
NoC,bb
|
07+12
|
71.6%
|
|
NoC,+EB
|
07+12
|
71.8%
|
|
NoC,+EB,bb
|
07+12
|
73.3%
|
Improving Object Detection with Deep Convolutional Networks via Bayesian Optimization and Structured Prediction
[td]
|
Model
|
BBoxReg?
|
VOC 2007 mAP(IoU>0.5)
|
|
R-CNN(AlexNet)
|
No
|
54.2%
|
|
R-CNN(VGG)
|
No
|
60.6%
|
|
+StructObj
|
No
|
61.2%
|
|
+StructObj-FT
|
No
|
62.3%
|
|
+FGS
|
No
|
64.8%
|
|
+StructObj+FGS
|
No
|
65.9%
|
|
+StructObj-FT+FGS
|
No
|
66.5%
|
[td]
|
Model
|
BBoxReg?
|
VOC 2007 mAP(IoU>0.5)
|
|
R-CNN(AlexNet)
|
Yes
|
58.5%
|
|
R-CNN(VGG)
|
Yes
|
65.4%
|
|
+StructObj
|
Yes
|
66.6%
|
|
+StructObj-FT
|
Yes
|
66.9%
|
|
+FGS
|
Yes
|
67.2%
|
|
+StructObj+FGS
|
Yes
|
68.5%
|
|
+StructObj-FT+FGS
|
Yes
|
68.4%
|
- arXiv: http://arxiv.org/abs/1504.03293
- slides: http://www.ytzhang.net/files/publications/2015-cvpr-det-slides.pdf
- code: https://github.com/YutingZhang/fgs-obj
Fast R-CNN
Fast R-CNN
[td]
|
method
|
data
|
VOC 2007 mAP
|
|
FRCN,VGG16
|
07
|
66.9%
|
|
FRCN,VGG16
|
07+12
|
70.0%
|
[td]
|
method
|
data
|
VOC 2010 mAP
|
|
FRCN,VGG16
|
12
|
66.1%
|
|
FRCN,VGG16
|
07++12
|
68.8%
|
[td]
|
method
|
data
|
VOC 2012 mAP
|
|
FRCN,VGG16
|
12
|
65.7%
|
|
FRCN,VGG16
|
07++12
|
68.4%
|
- arXiv: http://arxiv.org/abs/1504.08083
- slides: http://tutorial.caffe.berkeleyvision.org/caffe-cvpr15-detection.pdf
- github: https://github.com/rbgirshick/fast-rcnn
- webcam demo: https://github.com/rbgirshick/fast-rcnn/pull/29
- notes: http://zhangliliang.com/2015/05/17/paper-note-fast-rcnn/
- notes: http://blog.csdn.net/linj_m/article/details/48930179
- github(“Train Fast-RCNN on Another Dataset”): https://github.com/zeyuanxy/fast-rcnn/tree/master/help/train
DeepBox
DeepBox: Learning Objectness with Convolutional Networks
MR-CNN
Object detection via a multi-region & semantic segmentation-aware CNN model (MR-CNN)
[td]
|
Model
|
Trained on
|
VOC 2007 mAP
|
|
VGG-net
|
07+12
|
78.2%
|
|
VGG-net
|
07
|
74.9%
|
[td]
|
Model
|
Trained on
|
VOC 2012 mAP
|
|
VGG-net
|
07+12
|
73.9%
|
|
VGG-net
|
12
|
70.7%
|
- arXiv: http://arxiv.org/abs/1505.01749
- code: “Pdf and code will appear here shortly – stay tuned”
http://imagine.enpc.fr/~komodakn/ - notes: http://zhangliliang.com/2015/05/17/paper-note-ms-cnn/
- notes: http://blog.cvmarcher.com/posts/2015/05/17/multi-region-semantic-segmentation-aware-cnn/
Faster R-CNN
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks(NIPS 2015)
[td]
| training data |
test data
|
mAP
|
time/img
|
|
|
Faster RCNN, VGG-16
|
07
|
VOC 2007 test
|
69.9%
|
198ms
|
|
Faster RCNN, VGG-16
|
07+12
|
VOC 2007 test
|
73.2%
|
198ms
|
|
Faster RCNN, VGG-16
|
12
|
VOC 2007 test
|
67.0%
|
198ms
|
|
Faster RCNN, VGG-16
|
07++12
|
VOC 2007 test
|
70.4%
|
198ms
|
- arXiv: http://arxiv.org/abs/1506.01497
- github: https://github.com/ShaoqingRen/faster_rcnn
- github: https://github.com/rbgirshick/py-faster-rcnn
YOLO
You Only Look Once: Unified, Real-Time Object Detection(YOLO)
- intro: YOLO uses the whole topmost feature map to predict both confidences for multiple categories and bounding boxes (which are shared for these categories).
- arXiv: http://arxiv.org/abs/1506.02640
- code: http://pjreddie.com/darknet/yolo/
- github: https://github.com/pjreddie/darknet
- reddit:https://www.reddit.com/r/MachineLearning/comments/3a3m0o/realtime_object_detection_with_yolo/
- github(YOLO_tensorflow): https://github.com/gliese581gg/YOLO_tensorflow
R-CNN minus R
DenseBox
DenseBox: Unifying Landmark Localization with End to End Object Detection
- arXiv: http://arxiv.org/abs/1509.04874
- demo: http://pan.baidu.com/s/1mgoWWsS
- KITTI result: http://www.cvlibs.net/datasets/kitti/eval_object.php
SSD
SSD: Single Shot MultiBox Detector
- arXiv: http://arxiv.org/abs/1512.02325
- github: https://github.com/weiliu89/caffe/tree/ssd
- video: http://weibo.com/p/2304447a2326da963254c963c97fb05dd3a973
Inside-Outside Net
Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks
Detection results on VOC 2007 test:
[td]
|
Method
|
R
|
S
|
W
|
D
|
Train
|
mAP
|
|
FRCN
|
07+12
|
70.0
|
||||
|
RPN
|
07+12
|
73.2
|
||||
|
MR-CNN
|
√
|
07+12 |
78.2
|
|||
|
ION
|
07+12
|
74.6
|
||||
|
ION
|
√
|
07+12 |
75.6
|
|||
|
ION
|
√
|
√
|
07+12+S
|
76.5
|
||
|
ION
|
√
|
√
|
√
|
07+12+S |
78.5
|
|
|
ION
|
√
|
√
|
√
|
√
|
07+12+S
|
79.2
|
Detection results on VOC 2012 test:
[td]
|
Method
|
R
|
S
|
W
|
D
|
Train
|
mAP
|
|
FRCN
|
07++12
|
68.4
|
||||
|
RPN
|
07++12
|
70.4
|
||||
|
FRCN+YOLO
|
07++12
|
70.4
|
||||
|
HyperNet
|
07++12
|
71.4
|
||||
|
MR-CNN
|
√
|
07+12 |
73.9
|
|||
|
ION
|
√
|
√
|
√
|
√
|
07+12+S
|
76.4
|
- intro: “0.8s per image on a Titan X GPU (excluding proposal generation) without two-stage bounding-box regression and 1.15s per image with it”.
- arxiv: http://arxiv.org/abs/1512.04143
- slides: http://www.seanbell.ca/tmp/ion-coco-talk-bell2015.pdf
- coco-leaderboard: http://mscoco.org/dataset/#detections-leaderboard
G-CNN
G-CNN: an Iterative Grid Based Object Detector
Learning Deep Features for Discriminative Localization
- homepage: http://cnnlocalization.csail.mit.edu/
- arxiv: http://arxiv.org/abs/1512.04150
- github(Tensorflow): https://github.com/jazzsaxmafia/Weakly_detector
Factors in Finetuning Deep Model for object detection
We don’t need no bounding-boxes: Training object class detectors using only human verification
A MultiPath Network for Object Detection
Beyond Bounding Boxes: Precise Localization of Objects in Images (PhD Thesis)
- homepage: http://www.eecs.berkeley.edu/Pubs/TechRpts/2015/EECS-2015-193.html
- phd-thesis: http://www.eecs.berkeley.edu/Pubs/TechRpts/2015/EECS-2015-193.pdf
- github(“SDS using hypercolumns”): https://github.com/bharath272/sds
T-CNN: Tubelets with Convolutional Neural Networks for Object Detection from Videos
Training Region-based Object Detectors with Online Hard Example Mining
Specific Object Deteciton
End-to-end people detection in crowded scenes
- arXiv: http://arxiv.org/abs/1506.04878
- code: https://github.com/Russell91/reinspect
- ipn:http://nbviewer.ipython.org/github/Russell91/ReInspect/blob/master/evaluation_reinspect.ipynb
Tutorials
Convolutional Feature Maps: Elements of efficient (and accurate) CNN-based object detection
Codes
TensorBox: a simple framework for training neural networks to detect objects in images
- intro: “The basic model implements the simple and robust GoogLeNet-OverFeat algorithm. We additionally provide an implementation of the ReInspect algorithm”
- github: https://github.com/Russell91/TensorBox
Object detection in torch: Implementation of some object detection frameworks in torch
Blogs
Convolutional Neural Networks for Object Detection