Visual Planning on FROZENLAKE
91.6EM (%)VPRL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| VPRLInput=Image, Output=Image, post-trained=true, backbone=LVM-7B2025.05 | 91.6 | 93.2 | |
| VPFTInput=Image, Output=Image, post-trained=true, backbone=LVM-7B2025.05 | 75.4 | 79.5 | |
| Gemini 2.5 Pro (think)Input=Text + Image, Output=Text2025.05 | 72 | 85 | |
| Qwen 2.5-VL-Instruct-7B (SFT)Input=Text + Image, Output=Text, post-trained=true2025.05 | 68.6 | 84.4 | |
| Gemini 2.0 Flash (CoT)Input=Text + Image, Output=Text2025.05 | 27.6 | 52.5 | |
| Gemini 2.0 Flash (Direct)Input=Text + Image, Output=Text2025.05 | 21.2 | 47.6 | |
| Qwen 2.5-VL-Instruct-7B (CoT)Input=Text + Image, Output=Text2025.05 | 8.2 | 29.1 | |
| Qwen 2.5-VL-Instruct-7B (Direct)Input=Text + Image, Output=Text2025.05 | 1.2 | 15 |