Grounding on FigTalk-Gold Hard (reference set)
74.2Macro F1MINARD
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MINARDBackbone=Gemini2026.06 | 74.2 | 78.9 | 78.2 | |
| MINARDBackbone=GPT2026.06 | 66.5 | 72.6 | 70.8 | |
| MINARDBackbone=Claude2026.06 | 65.6 | 71.6 | 71.4 | |
| SAM Seg + BBoxBackbone=Claude2026.06 | 63.2 | 79.5 | 61.8 | |
| SAM Seg + BBoxBackbone=Gemini2026.06 | 57.2 | 77.2 | 55.1 | |
| VLM-GroundingBackbone=Gemini2026.06 | 56.6 | 62 | 60.1 | |
| VLM-GroundingBackbone=GPT2026.06 | 52.4 | 54.1 | 64.8 | |
| SAM Seg + BBoxBackbone=GPT2026.06 | 44.8 | 70.3 | 40.2 | |
| VLM-GroundingBackbone=Claude2026.06 | 41.4 | 51.8 | 40.7 |