Zero-shot Classification on CUB 2011 (test)
68.1Top-1 AccuracyReViSE
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| ReViSEtest data available during training=true2017.03 | 68.1 | — | — | — | 65.4 | 32.4 | 31.5 | |
| ReViSE†test data available during training=true, distributional matching=off2017.03 | 66.8 | — | — | — | 62.7 | 31.8 | 28.9 | |
| ReViSE††test data available during training=true, contractive loss=off, distributional matching=off2017.03 | 65.8 | — | — | — | 62.4 | 30 | 27.5 | |
| DeViSE*test data available during training=true2017.03 | 60.6 | — | — | — | 41.4 | 31.6 | 26.9 | |
| CMT*test data available during training=true2017.03 | 60.5 | — | — | — | 43.1 | 31.8 | 28.9 | |
| DS-SJEEmbedding=WORD CNN-RNN2016.05 | 56.8 | — | — | — | — | — | — | |
| DA-SJEEmbedding=WORD CNN-RNN2016.05 | 54.3 | — | — | — | — | — | — | |
| DA-SJEEmbedding=CHAR CNN-RNN2016.05 | 54 | — | — | — | — | — | — | |
| DS-SJEEmbedding=CHAR CNN-RNN2016.05 | 54 | — | — | — | — | — | — | |
| ReViSE♭test data available during training=true2017.03 | 53.6 | — | — | — | 53.7 | 26.4 | 28.2 | |
| DS-SJEEmbedding=WORD LSTM2016.05 | 53 | — | — | — | — | — | — | |
| DA-SJEEmbedding=WORD LSTM2016.05 | 52.2 | — | — | — | — | — | — | |
| DS-SJEEmbedding=WORD CNN2016.05 | 51 | — | — | — | — | — | — | |
| DA-SJEEmbedding=ATTRIBUTES2016.05 | 50.9 | — | — | — | — | — | — | |
| DA-SJEEmbedding=WORD CNN2016.05 | 50.5 | — | — | — | — | — | — | |
| CMTtest data available during training=false2017.03 | 50.5 | — | — | — | 42.4 | 29.6 | 25.7 | |
| DS-SJEEmbedding=ATTRIBUTES2016.05 | 50.4 | — | — | — | — | — | — | |
| ReViSETraining Data=labeled and unlabeled, Backbone=GoogLeNet, Attributes=att2017.03 | 49.6 | — | — | — | — | — | — | |
| DeViSEtest data available during training=false2017.03 | 49.3 | — | — | — | 40.8 | 28.8 | 25.6 | |
| DS-SJEEmbedding=CHAR CNN2016.05 | 48.2 | — | — | — | — | — | — | |
| ReViSEcTraining Data=labeled and unlabeled, Backbone=GoogLeNet, Attributes=att2017.03 | 48.2 | — | — | — | — | — | — | |
| ReViSEbTraining Data=labeled only, Backbone=GoogLeNet, Attributes=att2017.03 | 47.3 | — | — | — | — | — | — | |
| DA-SJEEmbedding=CHAR CNN2016.05 | 47.2 | — | — | — | — | — | — | |
| ReViSEaTraining Data=labeled only, Backbone=GoogLeNet, Attributes=att2017.03 | 45 | — | — | — | — | — | — | |
| DS-SJEEmbedding=BAG-OF-WORDS2016.05 | 44.1 | — | — | — | — | — | — | |
| DA-SJEEmbedding=BAG-OF-WORDS2016.05 | 43.4 | — | — | — | — | — | — | |
| CMTTraining Data=labeled only, Backbone=GoogLeNet, Attributes=att2017.03 | 40.6 | — | — | — | — | — | — | |
| CMT*Training Data=labeled and unlabeled, Backbone=GoogLeNet, Attributes=att2017.03 | 39.8 | — | — | — | — | — | — | |
| DA-SJEEmbedding=WORD2VEC2016.05 | 38.7 | — | — | — | — | — | — | |
| DS-SJEEmbedding=WORD2VEC2016.05 | 38.6 | — | — | — | — | — | — | |
| DeViSETraining Data=labeled only, Backbone=GoogLeNet, Attributes=att2017.03 | 36.8 | — | — | — | — | — | — | |
| DeViSE*Training Data=labeled and unlabeled, Backbone=GoogLeNet, Attributes=att2017.03 | 36.2 | — | — | — | — | — | — | |
| DA-SJEEmbedding=CHAR LSTM2016.05 | 22.6 | — | — | — | — | — | — | |
| DS-SJEEmbedding=CHAR LSTM2016.05 | 21.6 | — | — | — | — | — | — | |
| AHLEsupervision=supervised2014.09 | — | 27.3 | — | — | — | — | — | |
| ALEFeature Dimension=64K2014.09 | — | — | 22.3 | 26.9 | — | — | — | |
| SJEsupervision=unsupervised2014.09 | — | 29.9 | — | — | — | — | — | |
| SJEsupervision=supervised2014.09 | — | 51.7 | — | — | — | — | — | |
| SJEInput Embedding=Fisher Vectors (FV), Feature Dimension=40962014.09 | — | — | 15.2 | 19 | — | — | — | |
| SJEInput Embedding=Deep CNN Features (AlexNet), Feature Dimension=40962014.09 | — | — | 30 | 40.3 | — | — | — | |
| SJEInput Embedding=GoogLeNet (GOOG), Feature Dimension=10242014.09 | — | — | 37.8 | 50.1 | — | — | — |