Open-ended Visual Question Answering on VQA (test-standard)
83.3Accuracy (Overall)Human
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Humantype=Single Model2016.06 | 83.3 | 95.8 | 83.4 | 72.7 | |
| Ensemble of 7 Att. modelsTraining Split=train+val2016.06 | 66.5 | 83.2 | 39.5 | 58 | |
| Naver Labs (challenge 2nd)Training Split=train+val2016.06 | 64.8 | 83.3 | 38.7 | 54.6 | |
| Ours-ResNetBackbone=ResNet, type=Single Model2016.06 | 63.2 | 81.7 | 38.2 | 52.8 | |
| HieCoAttTraining Split=train+val2016.06 | 62.1 | — | — | — | |
| (Lu et al. 2016) ResNet*Backbone=ResNet, type=Single Model2016.06 | 62.1 | — | — | — | |
| Hierarchical Co-Attention (Alternating) + ResNetBackbone=ResNet, Attention Mechanism=Alternating co-attention2016.05 | 62.1 | — | — | — | |
| (Kim et al. 2016) ResNet*Backbone=ResNet, type=Single Model2016.06 | 61.8 | 82.4 | 38.2 | 49.4 | |
| DMN+Training Split=train+val2016.06 | 60.4 | — | — | — | |
| (Xiong, Merity, and Socher 2016)type=Single Model2016.06 | 60.4 | 80.4 | 36.8 | 48.3 | |
| DMN+Backbone=VGGNet2016.05 | 60.4 | — | — | — | |
| FDATraining Split=train+val2016.06 | 59.5 | — | — | — | |
| FDABackbone=ResNet2016.05 | 59.5 | — | — | — | |
| D-NMNTraining Split=train+val2016.06 | 59.4 | — | — | — | |
| AMATraining Split=train+val2016.06 | 59.4 | 81.1 | 37.1 | 45.8 | |
| (Wu et al. 2016)type=Single Model2016.06 | 59.4 | 81.1 | 37.1 | 45.8 | |
| SANTraining Split=train+val2016.06 | 58.9 | — | — | — | |
| (Yang et al. 2016)type=Single Model2016.06 | 58.9 | — | — | — | |
| SANBackbone=VGGNet2016.05 | 58.9 | — | — | — | |
| NMNTraining Split=train+val2016.06 | 58.7 | 81.2 | 37.7 | 44 | |
| AYNTraining Split=train+val2016.06 | 58.4 | 78.2 | 36.3 | 46.3 | |
| SMemTraining Split=train+val2016.06 | 58.2 | 80.9 | 37.5 | 43.5 | |
| VQA teamTraining Split=train+val2016.06 | 58.2 | 80.6 | 36.5 | 43.7 | |
| (Antol et al. 2015)type=Single Model2016.06 | 58.2 | 80.6 | 36.5 | 43.7 | |
| LSTM Q+IBackbone=VGGNet2016.05 | 58.2 | — | — | — | |
| SMemBackbone=GoogLeNet2016.05 | 58.2 | — | — | — | |
| (Andreas et al. 2016b)type=Single Model2016.06 | 58 | — | — | — | |
| DPPnetTraining Split=train+val2016.06 | 57.4 | 80.3 | 36.9 | 42.2 | |
| (Noh, Seo, and Han 2016)type=Single Model2016.06 | 57.4 | 80.3 | 36.9 | 42.2 | |
| iBOWIMGTraining Split=train+val2016.06 | 55.9 | 76.8 | 35 | 42.6 | |
| (Zhou et al. 2015)type=Single Model2016.06 | 55.9 | 76.8 | 35 | 42.6 | |
| (Andreas et al. 2016a)type=Single Model2016.06 | 55.1 | — | — | — |