Condition Rating on BridgeEQA (instances with < 30 images)
40.9Exact Match AccuracyEMVR VLM w/ SG Only
Evaluation Results
| Method | Links | |
|---|---|---|
| EMVR VLM w/ SG OnlyModel=Qwen3-VL 8B-Thinking, Note=Evaluated only on scenes with fewer than 30 images due to limited context window2025.11 | 40.9 | |
| Socratic LLM w/ SGModel=Qwen3-VL 8B-Thinking, Note=Evaluated only on scenes with fewer than 30 images due to limited context window2025.11 | 36.4 | |
| EMVR VLM w/ SG OnlyModel=Qwen3-VL 30B-A3B2025.11 | 29.3 | |
| EMVR VLM w/ SG OnlyModel=Nemotron-3 Nano-30B-A3B2025.11 | 23.2 | |
| Multi-Frame VLMModel=Nemotron-3 Nano-30B-A3B2025.11 | 11 | |
| Socratic LLM w/ SGModel=Nemotron-3 Nano-30B-A3B2025.11 | 11 | |
| Multi-Frame VLMModel=Qwen3-VL 8B-Thinking, Note=Evaluated only on scenes with fewer than 30 images due to limited context window2025.11 | 9.1 | |
| Multi-Frame VLMModel=Qwen3-VL 30B-A3B2025.11 | 8.5 | |
| Socratic LLM w/ SGModel=Qwen3-VL 30B-A3B2025.11 | 6.1 |