Match settings: 1000 categories of image classification problems, training data set 1.26 million images, validation set 50,000, test set 100,000 (callout not advertised). The data set is used by the 2012,2013,2014. The evaluation standard uses the TOP-5 error rate, that is, to predict an image 5 categories, as long as there is one and the same as the manual label category even if the right, otherwise wrong.
Score Leaderboard
Results announcement Time |
Institutions |
Top-5 error Rate (%) |
Number of models |
Method |
2015.2.6 |
MSRA |
4.94 |
|
http://arxiv.org/abs/1502.01852 |
2015.2.6 |
Baidu |
5.33 |
|
http://arxiv.org/abs/1501.02876 |
2015.1.13 |
Baidu |
5.98 |
|
—— |
2014.8.18 |
Google |
6.66 |
|
http://arxiv.org/abs/1409.4842 |
2014.8.18 |
Oxford |
7.33 |
|
http://arxiv.org/abs/1409.1556 |
2013.11.14 |
NYU |
11.7 |
|
http://arxiv.org/abs/1311.2901 |
2012.10.13 |
U.toronto |
16.4 |
|
Http://www.cs.toronto.edu/~fritz/absps/imagenet.pdf |
Continuous update, welcome comments to inform the latest results
Imagenet Image Classification Contest