Start-of-the-art Benchmark
COCO | Detection Leaderboard:
Timeline
Year |
Month |
Algorithm |
Improve |
Conference |
2012 |
|
AlexNet |
basemodel |
NIPS2012 |
= |
= |
= |
= |
= |
2013 |
11 |
R-CNN |
traditional method -> CNN method |
CVPR2014 |
2013 |
11 |
ZFNet |
basemodel |
|
= |
= |
= |
= |
= |
2014 |
6 |
SPPNet |
share feature extraction |
ECCV2014 |
2014 |
9 |
GoogleNet |
basemodel |
|
2014 |
9 |
VGG |
basemodel |
|
= |
= |
= |
= |
= |
2015 |
2 |
BN |
Easy to adjust parameters & accelerate convergence |
|
2015 |
4 |
Fast R-CNN |
4-stage -> 2-stage |
ICCV2015 |
2015 |
6 |
Faster R-CNN |
SS -> RPN |
NIPS2015 |
2015 |
6 |
YOLOv1 |
1-stage |
CVPR2016 |
2015 |
12 |
SSD |
multi-feature map |
ECCV2016 |
2015 |
12 |
ResNet |
basemodel |
CVPR2016 Best Paper |
= |
= |
= |
= |
= |
2016 |
5 |
R-FCN |
share sub-network |
NIPS2016 |
2016 |
8 |
DenseNet |
basemodel |
|
2016 |
10 |
Xception |
light-basemodel |
|
2016 |
11 |
ResNeXt |
basemodel |
|
2016 |
11 |
Speed/accuracy trade-offs |
Analyse |
|
2016 |
12 |
YOLOv2 |
multi-scale training、joint training |
CVPR2017 Best Paper Mention |
2016 |
12 |
FPN |
multi-feature map |
CVPR2017 |
= |
= |
= |
= |
= |
2017 |
1 |
DSSD |
SSD+FPN |
|
2017 |
3 |
Mask R-CNN |
RoIPooling -> RoIAlign |
ICCV2017 Best Paper |
2017 |
3 |
DCN |
Deformable module |
|
2017 |
4 |
Soft-NMS |
Eclectic version of NMS |
|
2017 |
4 |
MobileNet |
light-basemodel |
|
2017 |
7 |
ShuffleNet |
light-basemodel |
|
2017 |
7 |
DPN |
basemodel |
ImageNet2017 1st |
2017 |
8 |
RetinaNet |
CE -> FL |
ICCV2017 Best Student Paper |
2017 |
11 |
MegDet |
large-batch |
COCO2017 1st、CVPR2018 |
2017 |
11 |
Light-head R-CNN |
|
|
2017 |
11 |
Relation Network |
|
|
2017 |
11 |
SNIP |
image pyramid |
CVPR2018 |
2017 |
12 |
Cascade R-CNN |
Cascade module、2-stage -> 4-stage |
CVPR2018 |
2017 |
12 |
R-FCN-3000 |
large-scale detection |
|
= |
= |
= |
= |
= |
2018 |
3 |
PAN |
FPN -> PAN |
COCO2017 2nd |
2018 |
4 |
YOLOv3 |
|
|
2018 |
4 |
DetNet |
Detect-backbone |
|
2018 |
5 |
SNIPER |
|
|
2018 |
6 |
Understanding BN |
Analyse BN |
|