Image Captioning on Flickr30k (BLEU 1-4, PPL)
0.78BLEU-1Att-GT+LSTM
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Att-GT+LSTMAttribute Labels=Ground Truth2016.03 | 0.78 | 0.57 | 0.42 | 0.3 | 14.88 | |
| Att-RegionCNN+LSTMAttribute Labels=Predicted (Region CNN)2016.03 | 0.73 | 0.55 | 0.4 | 0.28 | 15.96 | |
| Att-GlobalCNN+LSTMAttribute Labels=Predicted (Global CNN)2016.03 | 0.7 | 0.5 | 0.35 | 0.27 | 16 | |
| Att-SVM+LSTMAttribute Labels=Predicted (SVM)2016.03 | 0.68 | 0.49 | 0.33 | 0.23 | 16.01 | |
| Xu et al. (Hard-Attention)Method Name=Hard-Attention2016.03 | 0.67 | 0.44 | 0.3 | 0.2 | — | |
| VggNet+ft+LSTMBackbone=VggNet, Feature Processing=Fine-tuning2016.03 | 0.67 | 0.47 | 0.31 | 0.21 | 16.62 | |
| Google(NIC)Method Name=NIC2016.03 | 0.66 | — | — | — | — | |
| Mao et al. (m-Rnn-VggNet)Backbone=VggNet, Method Name=m-Rnn2016.03 | 0.6 | 0.41 | 0.28 | 0.19 | 20.72 | |
| Donahue et al. (LRCN)Method Name=LRCN2016.03 | 0.59 | 0.39 | 0.25 | 0.17 | — | |
| VggNet-PCA+LSTMBackbone=VggNet, Feature Processing=PCA2016.03 | 0.59 | 0.4 | 0.26 | 0.17 | 18.92 | |
| Karpathy & Li (NeuralTalk)Method Name=NeuralTalk2016.03 | 0.57 | 0.37 | 0.24 | 0.16 | — | |
| VggNet+LSTMBackbone=VggNet, Feature Processing=Standard2016.03 | 0.57 | 0.38 | 0.25 | 0.17 | 18.83 | |
| Mao et al. (m-Rnn-AlexNet)Backbone=AlexNet, Method Name=m-Rnn2016.03 | 0.54 | 0.36 | 0.23 | 0.15 | 35.11 | |
| Chen & Zintick (Mind's Eye)Method Name=Mind's Eye2016.03 | — | — | — | 0.13 | 19.1 |