Semantic Grounding on COCO 2017 (val)
0.6719AccuracyOracle Verification (ours)
Evaluation Results
| Method | Links | |
|---|---|---|
| Oracle Verification (ours)VLM=GPT-4o, Dialogue round=t=32024.04 | 0.6719 | |
| Oracle Verification (ours)VLM=GPT-4o, Dialogue round=t=22024.04 | 0.6226 | |
| Oracle Verification (ours)VLM=GPT-4V, Dialogue round=t=32024.04 | 0.578 | |
| Oracle Verification (ours)VLM=GPT-4o, Dialogue round=t=12024.04 | 0.57 | |
| Oracle Verification (ours)VLM=GPT-4V, Dialogue round=t=22024.04 | 0.565 | |
| Oracle Verification (ours)VLM=GPT-4V, Dialogue round=t=12024.04 | 0.527 | |
| Oracle Verification (ours)VLM=ViP-LLaVA, Dialogue round=t=52024.04 | 0.5254 | |
| Oracle Verification (ours)VLM=ViP-LLaVA, Dialogue round=t=42024.04 | 0.5175 | |
| Oracle Verification (ours)VLM=ViP-LLaVA, Dialogue round=t=32024.04 | 0.5015 | |
| Oracle Verification (ours)VLM=LLaVA-1.5, Dialogue round=t=52024.04 | 0.4877 | |
| Oracle Verification (ours)VLM=ViP-LLaVA, Dialogue round=t=22024.04 | 0.4808 | |
| Intrinsic Self-CorrectionVLM=GPT-4o, Dialogue round=t=22024.04 | 0.4808 | |
| VLM Verification (ours)VLM=GPT-4o, Dialogue round=t=32024.04 | 0.4792 | |
| VLM Verification (ours)VLM=GPT-4o, Dialogue round=t=22024.04 | 0.4777 | |
| Oracle Verification (ours)VLM=LLaVA-1.5, Dialogue round=t=42024.04 | 0.4755 | |
| Intrinsic Self-CorrectionVLM=GPT-4o, Dialogue round=t=12024.04 | 0.4713 | |
| Intrinsic Self-CorrectionVLM=GPT-4o, Dialogue round=t=32024.04 | 0.4665 | |
| VLM Verification (ours)VLM=GPT-4o, Dialogue round=t=12024.04 | 0.4649 | |
| Oracle Verification (ours)VLM=LLaVA-1.5, Dialogue round=t=32024.04 | 0.4622 | |
| VLM Verification (ours)VLM=GPT-4V, Dialogue round=t=32024.04 | 0.4538 | |
| Oracle Verification (ours)VLM=ViP-LLaVA, Dialogue round=t=12024.04 | 0.449 | |
| VLM Verification (ours)VLM=GPT-4V, Dialogue round=t=22024.04 | 0.449 | |
| VLM Verification (ours)VLM=GPT-4V, Dialogue round=t=12024.04 | 0.4394 | |
| Oracle Verification (ours)VLM=LLaVA-1.5, Dialogue round=t=22024.04 | 0.4381 | |
| Oracle Verification (ours)VLM=LLaVA-1.5, Dialogue round=t=12024.04 | 0.4155 | |
| Intrinsic Self-CorrectionVLM=GPT-4V, Dialogue round=t=02024.04 | 0.4092 | |
| VLM Verification (ours)VLM=ViP-LLaVA, Dialogue round=t=52024.04 | 0.4044 | |
| VLM Verification (ours)VLM=ViP-LLaVA, Dialogue round=t=42024.04 | 0.4001 | |
| VLM Verification (ours)VLM=ViP-LLaVA, Dialogue round=t=22024.04 | 0.3964 | |
| VLM Verification (ours)VLM=ViP-LLaVA, Dialogue round=t=32024.04 | 0.3964 | |
| Intrinsic Self-CorrectionVLM=GPT-4o, Dialogue round=t=02024.04 | 0.3949 | |
| VLM Verification (ours)VLM=LLaVA-1.5, Dialogue round=t=52024.04 | 0.3821 | |
| VLM Verification (ours)VLM=ViP-LLaVA, Dialogue round=t=12024.04 | 0.3784 | |
| VLM Verification (ours)VLM=LLaVA-1.5, Dialogue round=t=42024.04 | 0.3769 | |
| Intrinsic Self-CorrectionVLM=ViP-LLaVA, Dialogue round=t=02024.04 | 0.3726 | |
| VLM Verification (ours)VLM=LLaVA-1.5, Dialogue round=t=32024.04 | 0.3704 | |
| VLM Verification (ours)VLM=LLaVA-1.5, Dialogue round=t=22024.04 | 0.3694 | |
| Intrinsic Self-CorrectionVLM=GPT-4V, Dialogue round=t=22024.04 | 0.3662 | |
| Intrinsic Self-CorrectionVLM=LLaVA-1.5, Dialogue round=t=02024.04 | 0.363 | |
| VLM Verification (ours)VLM=LLaVA-1.5, Dialogue round=t=12024.04 | 0.3587 | |
| Intrinsic Self-CorrectionVLM=LLaVA-1.5, Dialogue round=t=12024.04 | 0.3369 | |
| Intrinsic Self-CorrectionVLM=GPT-4V, Dialogue round=t=32024.04 | 0.328 | |
| Intrinsic Self-CorrectionVLM=ViP-LLaVA, Dialogue round=t=12024.04 | 0.3264 | |
| Intrinsic Self-CorrectionVLM=ViP-LLaVA, Dialogue round=t=22024.04 | 0.324 | |
| Intrinsic Self-CorrectionVLM=LLaVA-1.5, Dialogue round=t=22024.04 | 0.3226 | |
| Intrinsic Self-CorrectionVLM=LLaVA-1.5, Dialogue round=t=32024.04 | 0.3163 | |
| Intrinsic Self-CorrectionVLM=ViP-LLaVA, Dialogue round=t=32024.04 | 0.3112 | |
| Intrinsic Self-CorrectionVLM=GPT-4V, Dialogue round=t=12024.04 | 0.3089 | |
| Oracle Verification (ours)VLM=CogVLM, Dialogue round=t=52024.04 | 0.2125 | |
| Oracle Verification (ours)VLM=CogVLM, Dialogue round=t=42024.04 | 0.2101 | |
| Oracle Verification (ours)VLM=CogVLM, Dialogue round=t=32024.04 | 0.207 | |
| Oracle Verification (ours)VLM=CogVLM, Dialogue round=t=22024.04 | 0.2014 | |
| Oracle Verification (ours)VLM=CogVLM, Dialogue round=t=12024.04 | 0.1942 | |
| VLM Verification (ours)VLM=CogVLM, Dialogue round=t=52024.04 | 0.1884 | |
| VLM Verification (ours)VLM=CogVLM, Dialogue round=t=42024.04 | 0.1852 | |
| VLM Verification (ours)VLM=CogVLM, Dialogue round=t=32024.04 | 0.183 | |
| VLM Verification (ours)VLM=CogVLM, Dialogue round=t=22024.04 | 0.1783 | |
| VLM Verification (ours)VLM=CogVLM, Dialogue round=t=12024.04 | 0.1697 | |
| Intrinsic Self-CorrectionVLM=CogVLM, Dialogue round=t=22024.04 | 0.1647 | |
| Intrinsic Self-CorrectionVLM=CogVLM, Dialogue round=t=12024.04 | 0.1623 | |
| Intrinsic Self-CorrectionVLM=CogVLM, Dialogue round=t=32024.04 | 0.1592 | |
| Intrinsic Self-CorrectionVLM=CogVLM, Dialogue round=t=02024.04 | 0.148 |