Condition Rating on BridgeEQA fewer than 30 images
81.8Condition Rating Accuracy (±1)Multi-Frame VLM
Evaluation Results
| Method | Links | |
|---|---|---|
| Multi-Frame VLMVLM Backbone=Qwen3-VL 8B-Thinking2025.11 | 81.8 | |
| EMVR VLM w/ SG OnlyVLM Backbone=Qwen3-VL 8B-Thinking2025.11 | 81.8 | |
| EMVR VLM w/ SG OnlyVLM Backbone=Qwen3-VL 30B-A3B2025.11 | 76.8 | |
| Socratic LLM w/ SGVLM Backbone=Qwen3-VL 8B-Thinking2025.11 | 72.7 | |
| EMVR VLM w/ SG OnlyVLM Backbone=Nemotron-3 Nano-30B-A3B2025.11 | 70.7 | |
| Multi-Frame VLMVLM Backbone=Nemotron-3 Nano-30B-A3B2025.11 | 58.5 | |
| Socratic LLM w/ SGVLM Backbone=Nemotron-3 Nano-30B-A3B2025.11 | 48.8 | |
| Socratic LLM w/ SGVLM Backbone=Qwen3-VL 30B-A3B2025.11 | 30.5 | |
| Multi-Frame VLMVLM Backbone=Qwen3-VL 30B-A3B2025.11 | 23.2 |