Visual Question Answering on CLEVRER 1.0 (test)
0.94Descriptive AccuracyAloe
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| AloeObject Representation=MONet, Attention=Global cross-frame, Auxiliary Loss=Self-supervised2020.12 | 0.94 | 0.96 | 0.875 | 0.756 | — | — | |
| Aloe (No aux loss ablation)Auxiliary Loss=false2020.12 | 0.91 | 0.928 | 0.828 | 0.687 | — | — | |
| DCL2020.12 | 0.907 | 0.828 | 0.82 | 0.465 | — | — | |
| NS-DR2020.12 | 0.881 | 0.796 | 0.687 | 0.422 | — | — | |
| MAC (V+)2020.12 | 0.864 | 0.223 | 0.429 | 0.251 | — | — | |
| Aloe (Hierarchical ablation)Attention=hierarchical frame-level and intra-frame2020.12 | 0.806 | 0.874 | 0.735 | 0.551 | — | — | |
| Aloe (ResNet ablation)Object Representation=ResNet superpixels, Backbone=ResNet2020.12 | 0.749 | 0.661 | 0.583 | 0.324 | — | — | |
| Aloe (MLP ablation)Layers=4 fully connected layers, Units per layer=2048, Mechanism=MLP instead of self-attention2020.12 | 0.454 | 0.16 | 0.277 | 0.099 | — | — | |
| Aloe2025.06 | — | — | — | — | 0.914 | 0.756 | |
| CRCG_VRDPbase_method=VRDP2025.06 | — | — | — | — | 0.961 | 0.878 | |
| CRCG^approx_NSDRbase_method=NS-DR, approximation=true2025.06 | — | — | — | — | 0.907 | 0.783 | |
| DCL2025.06 | — | — | — | — | 0.804 | 0.465 | |
| NS-DR2025.06 | — | — | — | — | 0.741 | 0.422 | |
| ODDN-Aloe2025.06 | — | — | — | — | 0.93 | 0.801 | |
| VRDP2025.06 | — | — | — | — | 0.948 | 0.843 |