Visual Question Answering (Multiple-Choice) on VQA (test-dev)
69.4Overall Accuracy3-Modalities: Unary+Pairwise+Ternary
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| 3-Modalities: Unary+Pairwise+TernaryBackbone=ResNet2017.11 | 69.4 | 81.6 | 43.3 | 64.8 | |
| DANBackbone=ResNet2017.11 | 69.1 | — | — | — | |
| MLBBackbone=ResNet2017.11 | 68.9 | — | — | — | |
| 3-Modalities: Unary+PairwiseBackbone=ResNet2017.11 | 68.7 | 82 | 42.7 | 63.3 | |
| MCBBackbone=ResNet2017.11 | 68.6 | — | — | — | |
| 3-Modalities: Unary+Pairwise+TernaryBackbone=VGG2017.11 | 67.9 | 81.2 | 42.7 | 62.3 | |
| RAUBackbone=ResNet2017.11 | 67.7 | 81.9 | 41.1 | 61.5 | |
| DANBackbone=VGG2017.11 | 67 | — | — | — | |
| 2-Modalities: Unary+PairwiseBackbone=ResNet2017.11 | 66.7 | 80.9 | 36 | 61.6 | |
| Hierarchical Co-Attention (Alternating) + ResNetBackbone=ResNet, Attention Mechanism=Alternating co-attention2016.05 | 65.8 | 79.7 | 40 | 59.8 | |
| HieCoAttBackbone=ResNet2017.11 | 65.8 | 79.7 | 40 | 59.8 | |
| Hierarchical Co-Attention (Alternating) + VGGBackbone=VGGNet, Attention Mechanism=Alternating co-attention2016.05 | 64.9 | 79.7 | 40.1 | 57.9 | |
| HieCoAttBackbone=VGG2017.11 | 64.9 | 79.7 | 40.1 | 57.9 | |
| Hierarchical Co-Attention (Parallel) + VGGBackbone=VGGNet, Attention Mechanism=Parallel co-attention2016.05 | 64.6 | 79.5 | 39.8 | 57.4 | |
| FDABackbone=ResNet2016.05 | 64 | 81.5 | 39 | 54.7 | |
| LSTM Q+IBackbone=VGGNet2016.05 | 62.7 | 80.5 | 38.2 | 53 | |
| Region Sel.Backbone=VGGNet2016.05 | 62.4 | 77.6 | 34.3 | 55.8 |