Task success rate on DMV-Bench MC J=50 1.0
68.3Task Success RateDualMem
Evaluation Results
| Method | Links | |
|---|---|---|
| DualMemBack-end=Qwen2.5-VL-7B, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4072026.06 | 68.3 | |
| DualMemBack-end=Gemini 2.5 Flash, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4492026.06 | 65.1 | |
| M2ABack-end=Gemini 2.5 Flash, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4492026.06 | 64.7 | |
| CaptionBack-end=Qwen2.5-VL-7B, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4072026.06 | 58.8 | |
| M2ABack-end=Qwen2.5-VL-7B, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4072026.06 | 58.7 | |
| CaptionBack-end=Gemini 2.5 Flash, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4492026.06 | 47.7 | |
| WorldMMBack-end=Gemini 2.5 Flash, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4492026.06 | 37 | |
| MMABack-end=Gemini 2.5 Flash, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4492026.06 | 36.9 | |
| MMABack-end=Qwen2.5-VL-7B, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4072026.06 | 35.6 | |
| WorldMMBack-end=Qwen2.5-VL-7B, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4072026.06 | 30.9 | |
| NoMemoryBack-end=Qwen2.5-VL-7B, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4072026.06 | 0 | |
| TextOnlyBack-end=Qwen2.5-VL-7B, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4072026.06 | 0 | |
| NoMemoryBack-end=Gemini 2.5 Flash, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4492026.06 | 0 | |
| TextOnlyBack-end=Gemini 2.5 Flash, Monte Carlo range (r)=1–49, Probe counts (nr)=2,4492026.06 | 0 |