Spatial VQA on SQA3D (test)
50.3Overall AccuracySPATIOROUTE-R
Evaluation Results
| Method | Links | |
|---|---|---|
| SPATIOROUTE-RModel=Qwen3-2B, Prompt Strategy=Rule-based dynamic routing2026.05 | 50.3 | |
| SPATIOROUTE-LModel=Qwen3-2B, Prompt Strategy=LLM-driven prompt generation2026.05 | 49.9 | |
| SPATIOROUTE-RModel=Qwen2-7B, Prompt Strategy=Rule-based dynamic routing2026.05 | 49.5 | |
| SPATIOROUTE-LModel=Qwen2-7B, Prompt Strategy=LLM-driven prompt generation2026.05 | 48.8 | |
| BaselineModel=Qwen3-2B, Prompt Strategy=Fixed minimal prompt2026.05 | 48.3 | |
| CoTModel=Qwen3-2B, Prompt Strategy=Uniform step-by-step2026.05 | 47.7 | |
| BaselineModel=Qwen3-4B (thinking), Prompt Strategy=Fixed minimal prompt2026.05 | 47.7 | |
| BaselineModel=Qwen2-7B, Prompt Strategy=Fixed minimal prompt2026.05 | 47.3 | |
| SPATIOROUTE-RModel=Qwen2.5-3B, Prompt Strategy=Rule-based dynamic routing2026.05 | 47.1 | |
| SPATIOROUTE-RModel=Qwen2-2B, Prompt Strategy=Rule-based dynamic routing2026.05 | 46.7 | |
| Prior workModel=Qwen2-7B2026.05 | 46.5 | |
| SPATIOROUTE-LModel=Qwen2.5-3B, Prompt Strategy=LLM-driven prompt generation2026.05 | 46.5 | |
| SPATIOROUTE-LModel=Qwen2-2B, Prompt Strategy=LLM-driven prompt generation2026.05 | 45.7 | |
| BaselineModel=Qwen2.5-3B, Prompt Strategy=Fixed minimal prompt2026.05 | 44.8 | |
| CoTModel=Qwen2-7B, Prompt Strategy=Uniform step-by-step2026.05 | 42.1 | |
| BaselineModel=Qwen2-2B, Prompt Strategy=Fixed minimal prompt2026.05 | 42 | |
| CoTModel=Qwen2-2B, Prompt Strategy=Uniform step-by-step2026.05 | 38.9 | |
| CoTModel=Qwen2.5-3B, Prompt Strategy=Uniform step-by-step2026.05 | 38.5 | |
| CoTModel=Llama3.2-11B, Prompt Strategy=Uniform step-by-step2026.05 | 28.2 | |
| SPATIOROUTE-RModel=Llama3.2-11B, Prompt Strategy=Rule-based dynamic routing2026.05 | 24.2 | |
| SPATIOROUTE-LModel=Llama3.2-11B, Prompt Strategy=LLM-driven prompt generation2026.05 | 24 | |
| BaselineModel=Llama3.2-11B, Prompt Strategy=Fixed minimal prompt2026.05 | 23.3 |