Open-ended counting on TallyQA Complex 1.0 (test)
56.2Accuracy (ACC)RCN
Evaluation results
| Method | Links | ||
|---|---|---|---|
| RCNregion_proposals=Faster R-CNN2018.10 | 56.2 | 1.43 | |
| Zhang et al.region_proposals=Faster R-CNN2018.10 | 50.9 | 1.58 | |
| MUTANbackbone=ResNet-1522018.10 | 49.1 | 1.59 | |
| Q+Ibackbone=ResNet-152, architecture=MLP (1024 hidden units), modality=Image and Question2018.10 | 48.8 | 1.57 | |
| Guess-1description=Answer 1 for all questions2018.10 | 43.9 | 1.57 | |
| Q-Onlybackbone=RNN, architecture=MLP (1024 hidden units), modality=Question only2018.10 | 39.1 | 1.75 | |
| I-Onlybackbone=ResNet-152, architecture=MLP (1024 hidden units), modality=Image only2018.10 | 26.4 | 1.69 | |
| Guess-2description=Answer 2 for all questions2018.10 | 15.9 | 1.69 | |
| DETECTbackbone=YOLOv22018.10 | 15 | 4.52 |