Object Hallucination Assessment on AMBER
16.2CHAIR_IHALC
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HALCLVLM Backbone=Shikra, Maximum generation length=256 and 512 tokens2026.03 | 16.2 | 39.6 | |
| DML-LVLMLVLM Backbone=InstructBLIP, Maximum generation length=256 and 512 tokens2026.03 | 13.6 | 44.3 | |
| HALCLVLM Backbone=InstructBLIP, Maximum generation length=256 and 512 tokens2026.03 | 13.1 | 49.3 | |
| EAHLVLM Backbone=Shikra, Maximum generation length=256 and 512 tokens2026.03 | 12.6 | 52.3 | |
| OPERALVLM Backbone=InstructBLIP, Maximum generation length=256 and 512 tokens2026.03 | 12.2 | 48.7 | |
| OPERALVLM Backbone=LLaVA-1.5, Maximum generation length=256 and 512 tokens2026.03 | 12.1 | 45.3 | |
| EAHLVLM Backbone=LLaVA-1.5, Maximum generation length=256 and 512 tokens2026.03 | 11.3 | 48.4 | |
| OPERALVLM Backbone=Shikra, Maximum generation length=256 and 512 tokens2026.03 | 11.2 | 42.6 | |
| GIFTLVLM Backbone=InstructBLIP, Maximum generation length=256 and 512 tokens2026.03 | 11.2 | 39.4 | |
| HALCLVLM Backbone=MiniGPT-4, Maximum generation length=256 and 512 tokens2026.03 | 11.1 | 53.6 | |
| EAHLVLM Backbone=InstructBLIP, Maximum generation length=256 and 512 tokens2026.03 | 10.7 | 49.1 | |
| EAHLVLM Backbone=MiniGPT-4, Maximum generation length=256 and 512 tokens2026.03 | 10.6 | 54.5 | |
| OPERALVLM Backbone=MiniGPT-4, Maximum generation length=256 and 512 tokens2026.03 | 10.3 | 50.4 | |
| GIFTLVLM Backbone=Shikra, Maximum generation length=256 and 512 tokens2026.03 | 10.2 | 47.2 | |
| HALCLVLM Backbone=LLaVA-1.5, Maximum generation length=256 and 512 tokens2026.03 | 9.7 | 51.7 | |
| LVLMs-SaliencyLVLM Backbone=Shikra, Maximum generation length=256 and 512 tokens2026.03 | 9.6 | 50.8 | |
| GIFTLVLM Backbone=MiniGPT-4, Maximum generation length=256 and 512 tokens2026.03 | 9.6 | 41.8 | |
| LVLMs-SaliencyLVLM Backbone=InstructBLIP, Maximum generation length=256 and 512 tokens2026.03 | 9.6 | 46.9 | |
| DML-LVLMLVLM Backbone=Shikra, Maximum generation length=256 and 512 tokens2026.03 | 9.2 | 46.7 | |
| GIFTLVLM Backbone=LLaVA-1.5, Maximum generation length=256 and 512 tokens2026.03 | 8.9 | 50.6 | |
| DML-LVLMLVLM Backbone=MiniGPT-4, Maximum generation length=256 and 512 tokens2026.03 | 8.9 | 47.4 | |
| OPERALVLM Backbone=Qwen2-VL-7B, Maximum generation length=256 and 512 tokens2026.03 | 8.8 | 52.4 | |
| LVLMs-SaliencyLVLM Backbone=MiniGPT-4, Maximum generation length=256 and 512 tokens2026.03 | 8.4 | 56.7 | |
| SAGELVLM Backbone=Shikra, Maximum generation length=256 and 512 tokens2026.03 | 8.3 | 53.4 | |
| SAGELVLM Backbone=InstructBLIP, Maximum generation length=256 and 512 tokens2026.03 | 8.2 | 55.7 | |
| DML-LVLMLVLM Backbone=Qwen2-VL-7B, Maximum generation length=256 and 512 tokens2026.03 | 8.1 | 51.6 | |
| LVLMs-SaliencyLVLM Backbone=LLaVA-1.5, Maximum generation length=256 and 512 tokens2026.03 | 7.9 | 51.3 | |
| DML-LVLMLVLM Backbone=LLaVA-1.5, Maximum generation length=256 and 512 tokens2026.03 | 7.8 | 52.6 | |
| HALCLVLM Backbone=Qwen2-VL-7B, Maximum generation length=256 and 512 tokens2026.03 | 7.8 | 54.7 | |
| SAGELVLM Backbone=MiniGPT-4, Maximum generation length=256 and 512 tokens2026.03 | 7.6 | 53.2 | |
| EAHLVLM Backbone=Qwen2-VL-7B, Maximum generation length=256 and 512 tokens2026.03 | 7.1 | 55.2 | |
| SAGELVLM Backbone=LLaVA-1.5, Maximum generation length=256 and 512 tokens2026.03 | 6.9 | 57.2 | |
| LVLMs-SaliencyLVLM Backbone=Qwen2-VL-7B, Maximum generation length=256 and 512 tokens2026.03 | 6.7 | 64.9 | |
| GIFTLVLM Backbone=Qwen2-VL-7B, Maximum generation length=256 and 512 tokens2026.03 | 6.3 | 58.7 | |
| SAGELVLM Backbone=Qwen2-VL-7B, Maximum generation length=256 and 512 tokens2026.03 | 6.1 | 61.5 |