Question Answering on GVQA v1.0 (test)
43.89SPICEKLDrive-GT
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| KLDrive-GTTraining Protocol=use frozen LLM with few-shot ICL only, variant=KG from ground-truth annotations2026.03 | 43.89 | 0.43 | 3.88 | |
| LLaVA-v1.6-mistral-7bTraining Protocol=use frozen LLM with few-shot ICL only2026.03 | 43.75 | 0.11 | 0.38 | |
| KLDrive-KGTraining Protocol=use frozen LLM with few-shot ICL only, variant=full system with predicted perception outputs2026.03 | 42.45 | 0.41 | 3.48 | |
| DriveLMTraining Protocol=trained since they have task-specific parameters2026.03 | 42.16 | 0.42 | 3.53 | |
| LiDAR-LLMTraining Protocol=trained since they have task-specific parameters2026.03 | 41.09 | 0.42 | 3.35 | |
| CREMATraining Protocol=trained since they have task-specific parameters2026.03 | 39.05 | 0.39 | 3.27 | |
| MAPLMTraining Protocol=trained since they have task-specific parameters2026.03 | 37.12 | 0.36 | 3.09 | |
| Qwen3Vision-8BTraining Protocol=use frozen LLM with few-shot ICL only2026.03 | 36.44 | 0.22 | 0.5 | |
| Internvl-3.5-8bTraining Protocol=use frozen LLM with few-shot ICL only2026.03 | 32.66 | 0.13 | 0.34 | |
| FocalFormer3DTraining Protocol=trained since they have task-specific parameters2026.03 | 14.41 | 0.21 | 0.56 | |
| RayDNTraining Protocol=trained since they have task-specific parameters2026.03 | 11.58 | 0.2 | 0.46 | |
| IS-FusionTraining Protocol=trained since they have task-specific parameters2026.03 | 11.2 | 0.19 | 0.44 |