ILSVRC 2014结束一段时间了。从下面的表格来看，基本都是RCNN的路子，但是这些牛队都做了改进。自己和人家比差的太远啊，努力。
|
team |
results |
Spotlights and improve |
|
GoogLeNet |
0.439329(6 m) 0.38(1m) |
Rcnn 1. Increase size of super-pixels by 2X 2. Add multibox* proposals |
|
CUHK DeepID-Net |
0.406659 |
RCNN + Bounding box rejection using def-pooling layer 1000 object-level annotation 200 object-level annotation |
|
Deep Insight |
0.404517 |
Original RCNN + 9conv + SPM + more iterations + Structural Edge Proposal + 7/8/9 Conv Ensemble + CLS Context |
|
NUS |
0.37212 |
Rcnn framework, with nin in cnn |
|
UvA-Euvision |
0.354213(aug) 0.32.253(prov) |
Selective search + cnn |
|
MSRA Visual Computing |
0.351103 |
A combination of multiple SPP-net-based models (no outside data) |
|
Berkeley Vision |
0.345213 |
R-CNN baseline |
原文：http://www.cnblogs.com/jianyingzhou/p/3998210.html