3D Question Answering on NuscenesQA v1.0 (test)
85.4Existence Accuracy (All)MSMDFusion + MCAN
Evaluation results
| Method | Links | ||||||||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MSMDFusion + MCANBackbone=MSMDFusion, QA Head=MCAN, Input Modality=LiDAR and camera2025.12 | 85.4 | — | — | — | — | 22.2 | — | — | 60.6 | — | — | 69.7 | — | — | 54.3 | 60.4 | — | — | — | — | — | — | |
| MSMDFusion + BUTDBackbone=MSMDFusion, QA Head=BUTD, Input Modality=LiDAR and camera2025.12 | 85.1 | — | — | — | — | 23.2 | — | — | 59.5 | — | — | 68.5 | — | — | 52.3 | 59.8 | — | — | — | — | — | — | |
| CenterPoint+MCANPerception Backbone=CenterPoint, VQA Architecture=MCAN2025.12 | 84.8 | 87.7 | 82.3 | 22.5 | 19.1 | 20.8 | 66.6 | 56.3 | 59.8 | 82.4 | 68.8 | 70 | 71.3 | 49 | 52.3 | 59.5 | — | — | — | — | — | — | |
| CenterPoint + MCANBackbone=CenterPoint, QA Head=MCAN, Input Modality=LiDAR-only2025.12 | 84.8 | — | — | — | — | 20.8 | — | — | 59.8 | — | — | 70 | — | — | 52.3 | 59.5 | — | — | — | — | — | — | |
| BEVDet+MCANPerception Backbone=BEVDet, VQA Architecture=MCAN2025.12 | 84.2 | 87.2 | 81.7 | 21.8 | 19.2 | 20.4 | 64.1 | 49.9 | 54.7 | 75.1 | 66.7 | 67.4 | 73 | 47.4 | 51.2 | 57.9 | — | — | — | — | — | — | |
| BEVDet + MCANBackbone=BEVDet, QA Head=MCAN, Input Modality=Multi-camera2025.12 | 84.2 | — | — | — | — | 20.4 | — | — | 54.7 | — | — | 67.4 | — | — | 51.2 | 57.9 | — | — | — | — | — | — | |
| NuscenesQA*Specialist Model=true2024.12 | 84.1 | 87.7 | 81.1 | 21.9 | 20.7 | 21.3 | 70.2 | 45.6 | 49.2 | 62.8 | 52.4 | 55.9 | 81.6 | 68 | 69.2 | 58.1 | — | — | — | — | — | — | |
| CenterPoint+BUTDPerception Backbone=CenterPoint, VQA Architecture=BUTD2025.12 | 84.1 | 87.7 | 81.1 | 21.9 | 20.7 | 21.3 | 62.8 | 52.4 | 55.9 | 81.6 | 68 | 69.2 | 70.2 | 45.6 | 49.2 | 58.1 | — | — | — | — | — | — | |
| CenterPoint + BUTDBackbone=CenterPoint, QA Head=BUTD, Input Modality=LiDAR-only2025.12 | 84.1 | — | — | — | — | 21.3 | — | — | 55.9 | — | — | 69.2 | — | — | 49.2 | 58.1 | — | — | — | — | — | — | |
| MMDriveLLM Backbone=T5, Vision Encoder=UniRepLKNet, Occupancy Encoding=UniScene2025.12 | 83.9 | 86.7 | 81.6 | 28.1 | 30.3 | 29.2 | 69.3 | 60.5 | 63.5 | 85.4 | 74.4 | 75.3 | 72.1 | 51.2 | 54.3 | 62.7 | — | — | — | — | — | — | |
| BEVDet+BUTDPerception Backbone=BEVDet, VQA Architecture=BUTD2025.12 | 83.7 | 87.2 | 80.6 | 21.7 | 20 | 20.9 | 55 | 50.5 | 52 | 76.1 | 66.8 | 67.7 | 69.4 | 45.2 | 48.8 | 57 | — | — | — | — | — | — | |
| BEVDet + BUTDBackbone=BEVDet, QA Head=BUTD, Input Modality=Multi-camera2025.12 | 83.7 | — | — | — | — | 20.9 | — | — | 51.3 | — | — | 66.9 | — | — | 42 | 53.4 | — | — | — | — | — | — | |
| LSceneLLM2024.12 | 83.6 | 86.4 | 81.3 | 19.4 | 19.8 | 19.6 | 64.4 | 41.3 | 44.8 | 58.8 | 51.2 | 53.8 | 81 | 67.5 | 68.7 | 56.4 | — | — | — | — | — | — | |
| BeLLA w/ LLamaBackbone=BeLLA, LLM Backbone=LLama 3.2B, Input Modality=Multi-camera2025.12 | 80.9 | — | — | — | — | 20.5 | — | — | 69.9 | — | — | 62.7 | — | — | 54.8 | 59.6 | — | — | — | — | — | — | |
| OccLLaMA2024.12 | 79.9 | 80.6 | 79.3 | 18.6 | 19.1 | 18.9 | 64.9 | 39 | 42.8 | 48 | 49.6 | 49.1 | 80.6 | 63.7 | 65.2 | 53.4 | — | — | — | — | — | — | |
| Q-OnlyInput=Question only2025.12 | 79.6 | 81.7 | 77.9 | 17.8 | 16.5 | 17.2 | 57.2 | 48.3 | 51.3 | 79.5 | 65.7 | 66.9 | 59.4 | 38.9 | 42 | 53.4 | — | — | — | — | — | — | |
| LidarLLM2024.12 | 74.5 | 79.1 | 70.6 | 15.3 | 14.7 | 15 | 59.6 | 34.1 | 37.8 | 53.4 | 42 | 45.9 | 67 | 57 | 57.8 | 48.6 | — | — | — | — | — | — | |
| LLaVA2024.12 | 45.8 | 38.9 | 51.9 | 7.7 | 7.6 | 7.7 | 10.5 | 7.4 | 7.8 | 2 | 9.9 | 9 | 64.5 | 50.8 | 52.1 | 26.2 | — | — | — | — | — | — | |
| LLaVA-Adaptaer-v22024.12 | 19.3 | 34.2 | 6.3 | 5 | 0.1 | 2.7 | 23.7 | 4.6 | 7.6 | 9.8 | 11.3 | 10.8 | 2.6 | 1.5 | 1.6 | 9.6 | — | — | — | — | — | — | |
| BEVDet+BUTD2025.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 83.7 | 22 | 48.8 | 52 | 67.7 | 57 | |
| LiDAR-LLM2025.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 74.5 | 15 | 37.8 | 45.9 | 57.8 | 48.6 | |
| LLaMA-AdapV22025.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 19.3 | 2.7 | 7.6 | 10.8 | 1.6 | 9.6 | |
| LLaVA1.52025.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 45.8 | 7.7 | 7.8 | 9 | 52.1 | 26.2 | |
| OccLLaMA3.12025.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 80.9 | 19.2 | 46.3 | 47.8 | 66.6 | 54.5 | |
| OpenDriveVLAParameters=0.5B2025.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 83.9 | 22 | 50.2 | 57 | 68.4 | 58.4 | |
| OpenDriveVLAParameters=3B2025.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 84 | 22.3 | 50.3 | 56.9 | 68.5 | 58.5 | |
| OpenDriveVLAParameters=7B2025.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 84.2 | 22.7 | 49.6 | 54.5 | 68.8 | 58.2 | |
| SNOWtraining-free=true2025.12 | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | — | 82.3 | 27.4 | 53.2 | 80.5 | 61 | 60.1 |