Visual Grounding on RefCOCO (test)
91.4AccuracyOriginal Grounding CogVLM
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Original Grounding CogVLM2024.06 | 91.4 | 0 | |
| Non-merging CogVLM2024.06 | 91.1 | 0 | |
| HPE-CogVLM2024.06 | 90.5 | 0 | |
| OptMerge2025.05 | 89.94 | — | |
| Individual VQA2025.05 | 89.74 | — | |
| Individual Chart2025.05 | 89.72 | — | |
| Individual OCR2025.05 | 89.63 | — | |
| Individual Geometry2025.05 | 89.58 | — | |
| TA merging CogVLMlambda=0.52024.06 | 89.5 | 0 | |
| AutoToolSize=7B2026.05 | 88.5 | — | |
| Individual Grounding2025.05 | 88.32 | — | |
| Qwen2.5-VL-32B-Instruct2025.05 | 88.01 | — | |
| DeepEyesSize=7B2026.05 | 86 | — | |
| Qwen2.5-VL-7BSize=7B2026.05 | 84.7 | — | |
| LLaVAAvg Tokens=5762026.02 | 58.3 | — | |
| PoolingAvg Tokens=1282026.02 | 23.01 | — | |
| PoolingAvg Tokens=642026.02 | 12.01 | — | |
| FastVAvg Tokens=1282026.02 | 10.34 | — | |
| SparseVLMAvg Tokens=1282026.02 | 6.27 | — | |
| VisionZipAvg Tokens=1282026.02 | 4.49 | — | |
| VisionZipAvg Tokens=642026.02 | 4.04 | — | |
| FastVAvg Tokens=642026.02 | 2.73 | — | |
| SparseVLMAvg Tokens=642026.02 | 1.04 | — |