3D Visual Question Answering on 3DMV-VQA (held-out)
68.9Concept Accuracy3D-LLM
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| 3D-LLMBackbone=flamingo2023.07 | 68.9 | 32.4 | 61.6 | 68.3 | 58.6 | |
| 3D-LLMBackbone=BLIP2-flanT52023.07 | 68.1 | 31.4 | 55.1 | 69.7 | 54.6 | |
| 3D-CLRexplicit object representations and neuro-symbolic reasoning=true2023.07 | 66.1 | 41.3 | 57.6 | 72.3 | 57.7 | |
| 3D-LLMBackbone=BLIP2-opt2023.07 | 63.4 | 30.7 | 57.6 | 65.2 | 54.9 | |
| BLIPInput=MultiView2023.07 | 61.9 | 21.1 | 48 | 62.3 | 47.1 | |
| 3D-Feature+LSTM2023.07 | 61.2 | 22.4 | 49.9 | 61.3 | 48.2 | |
| flamingoInput=MultiView2023.07 | 60 | 18.3 | 40.2 | 61.4 | 41.6 | |
| NS-VQAexplicit object representations and neuro-symbolic reasoning=true2023.07 | 59.8 | 21.5 | 33.4 | 61.6 | 38 | |
| flamingoInput=SingleImage2023.07 | 58.7 | 18.5 | 38.4 | 60.1 | 40.3 | |
| BLIPInput=SingleImage2023.07 | 58 | 20.4 | 42.3 | 62.3 | 43.1 |