Generative Hallucination on AMBER Generative
70.4Coverage (%)Qwen2.5-VL 3B + SFT
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Qwen2.5-VL 3B + SFTModel=Qwen2.5-VL 3B, Training Strategy=+ SFT2025.10 | 70.4 | — | 48.8 | 3.9 | 6.9 | — | — | |
| Qwen2.5-VL 3BModel=Qwen2.5-VL 3B, Training Strategy=Base2025.10 | 69.9 | — | 55.7 | 5.8 | 7.6 | — | — | |
| Qwen2.5-VL 3B + GRPOModel=Qwen2.5-VL 3B, Training Strategy=+ GRPO2025.10 | 68.5 | — | 44.6 | 4.2 | 6.7 | — | — | |
| Qwen2.5-VL 3B + NoisyGRPOModel=Qwen2.5-VL 3B, Training Strategy=+ NoisyGRPO2025.10 | 67.7 | — | 44.3 | 3.4 | 6.6 | — | — | |
| GPT-4V (Base)Backbone=GPT-4V2025.03 | 67.1 | 4.6 | 30.7 | 2.6 | — | — | — | |
| GPT4VModel=GPT4V2025.10 | 67.1 | — | 30.7 | 2.6 | 4.6 | — | — | |
| Qwen2.5-VL 7B + SFTModel=Qwen2.5-VL 7B, Training Strategy=+ SFT2025.10 | 64.9 | — | 32.4 | 2.1 | 5.4 | — | — | |
| Qwen2.5-VL 7B + GRPOModel=Qwen2.5-VL 7B, Training Strategy=+ GRPO2025.10 | 64.3 | — | 28.9 | 1.7 | 4.9 | — | — | |
| Qwen2.5-VL 7BModel=Qwen2.5-VL 7B, Training Strategy=Base2025.10 | 63.6 | — | 27.5 | 1.6 | 4.8 | — | — | |
| Qwen2.5-VL 7B + NoisyGRPOModel=Qwen2.5-VL 7B, Training Strategy=+ NoisyGRPO2025.10 | 63.2 | — | 23.9 | 1.2 | 4.2 | — | — | |
| CogVLM-ChatModel=CogVLM-Chat2025.10 | 57.2 | — | 23.6 | 1.3 | 5.6 | — | — | |
| VCDBase Model=LLaVA-Next-7B2026.04 | 56.1 | 7.1 | 36.3 | 3 | — | — | — | |
| LLaVA-Next-7BBase Model=LLaVA-Next-7B2026.04 | 55.8 | 7.1 | 37.6 | 3.1 | — | — | — | |
| M3IDBase Model=LLaVA-Next-7B2026.04 | 55.6 | 6.9 | 35.8 | 3.3 | — | — | — | |
| AVISCBase Model=LLaVA-Next-7B2026.04 | 55.2 | 7.5 | 38 | 2.5 | — | — | — | |
| AttnRealBase Model=InstructBLIP2026.05 | 54.3 | 5.7 | 30.5 | 2.9 | — | 27.2 | — | |
| AttnRealBase Model=Qwen-VL2026.05 | 54.2 | 5.3 | 25.9 | 2.1 | — | 0.7 | — | |
| VCDBase Model=InstructBLIP2026.05 | 54.2 | 8.3 | 39.1 | 3.7 | — | 2.1 | — | |
| VanillaBase Model=InstructBLIP2026.05 | 53.8 | 8.1 | 35.3 | 3.7 | — | 2.3 | — | |
| VanillaBase Model=Qwen-VL2026.05 | 53.3 | 6.3 | 29.6 | 2.6 | — | 0.2 | — | |
| HALVABackbone=LLaVA-v1.5-7B2025.03 | 53 | 6.6 | 32.2 | 3.4 | — | — | — | |
| VCD + CNSenhancement=Contrastive Neuron Steering2026.01 | 52.8 | 6.9 | 34.2 | 3.7 | — | — | — | |
| Vanilla + CNSenhancement=Contrastive Neuron Steering2026.01 | 52.4 | 7.1 | 34.7 | 3.9 | — | — | — | |
| VCDBase Model=Shikra2026.05 | 52.3 | 10.7 | 50 | 4.9 | — | 0.2 | — | |
| PSRDBase Model=LLaVA-1.5-13B2026.04 | 52 | 4.7 | 24.1 | 2.2 | — | — | — | |
| CASBase Model=InstructBLIP2026.05 | 51.9 | 6.1 | 28.8 | 2.8 | — | 2.6 | — | |
| PADE2026.02 | 51.8 | 7.1 | 31.4 | 3.4 | — | — | — | |
| AttnRealBase Model=Shikra2026.05 | 51.7 | 8.7 | 43.3 | 4.3 | — | 0.5 | — | |
| DoLA2026.02 | 51.6 | 7.6 | 36 | 4 | — | — | — | |
| VanillaBase Model=Shikra2026.05 | 51.4 | 10.8 | 48.5 | 5.5 | — | 0.3 | — | |
| Vanilla2026.01 | 51 | 7.8 | 36.4 | 4.2 | — | — | — | |
| Vanilla2026.02 | 51 | 7.8 | 36.4 | 4.2 | — | — | — | |
| LLaVA-1.5 7BModel=LLaVA-1.5 7B2025.10 | 51 | — | 36.4 | 4.2 | 7.8 | — | — | |
| VCD2026.01 | 50.8 | 7.4 | 36.1 | 4 | — | — | — | |
| VCD2026.02 | 50.8 | 7.5 | 36.2 | 4.1 | — | — | — | |
| VCDBase Model=Qwen-VL2026.05 | 50.8 | 8.8 | 37.5 | 3.4 | — | 0.2 | — | |
| PAIBase Model=LLaVA-1.52026.05 | 50.5 | 4.4 | 23.2 | 1.6 | — | 29 | — | |
| CASBase Model=Shikra2026.05 | 50.5 | 7.6 | 34.2 | 2.9 | — | 0.1 | — | |
| VCDBase Model=LLaVA-1.52026.05 | 50.4 | 8.7 | 36.2 | 3.9 | — | 0.5 | — | |
| AttnRealBase Model=LLaVA-1.52026.05 | 50.2 | 5.3 | 26 | 2.4 | — | 4.9 | — | |
| CASBase Model=Qwen-VL2026.05 | 50.2 | 5.3 | 23.5 | 1.9 | — | 0.5 | — | |
| PAI2026.02 | 49.9 | 7.4 | 33.2 | 3.7 | — | — | — | |
| HA-DPOBackbone=LLaVA-v1.5-7B2025.03 | 49.8 | 6.7 | 30.9 | 3.3 | — | — | — | |
| OPERA2026.02 | 49.6 | 7.3 | 32 | 3.5 | — | — | — | |
| AVISCBase Model=LLaVA-1.5-13B2026.04 | 49.5 | 8.6 | 39.6 | 3.6 | — | — | — | |
| Qwen-VL-ChatModel=Qwen-VL-Chat2025.10 | 49.4 | — | 23.6 | 1.9 | 5.5 | — | — | |
| LLaVA-1.5-13BBase Model=LLaVA-1.5-13B2026.04 | 49.4 | 6.7 | 28.8 | 3.1 | — | — | — | |
| VanillaBase Model=LLaVA-1.52026.05 | 49.4 | 7.7 | 31.1 | 3.6 | — | 0.5 | — | |
| OctopusBackbone=LLaVA-v1.5-7B2025.03 | 49.2 | 4.8 | 23.4 | 1.2 | — | — | — | |
| EOSBackbone=LLaVA-v1.5-7B2025.03 | 49.1 | 5.1 | 22.7 | 2 | — | — | — | |
| M3IDBackbone=LLaVA-v1.5-7B, reproduced=true2025.03 | 48.9 | 6 | 26 | 1.5 | — | — | — | |
| YARDBackbone=LLaVA-1.5-7B2026.05 | 48.9 | 4.6 | 21.6 | 1.2 | — | — | — | |
| M3IDBackbone=LLaVA-1.5-7B2026.05 | 48.7 | 5.4 | 25.5 | 1.4 | — | — | — | |
| OctopusBackbone=Instruct-BLIP2025.03 | 48.5 | 6.1 | 22.2 | 1.3 | — | — | — | |
| OctopusBase Model=InstructBLIP-7B2026.04 | 48.5 | 6.1 | 22.2 | 1.3 | — | — | — | |
| VCDBase Model=LLaVA-1.5-13B2026.04 | 48.3 | 6.7 | 33.6 | 3.5 | — | — | — | |
| FuzzyCDBackbone=LLaVA-1.5-7B2026.05 | 48.1 | 6.2 | 27.3 | 1.8 | — | — | — | |
| PSRDBase Model=InstructBLIP-7B2026.04 | 47.8 | 4.4 | 20.9 | 1.8 | — | — | — | |
| VCDBackbone=Instruct-BLIP, reproduced=true2025.03 | 47.7 | 7.6 | 29.9 | 2.2 | — | — | — | |
| VCDBase Model=InstructBLIP-7B2026.04 | 47.7 | 7.6 | 29.9 | 2.2 | — | — | — | |
| M3IDBase Model=LLaVA-1.5-13B2026.04 | 47.7 | 6 | 30.4 | 3.3 | — | — | — | |
| M3IDBackbone=Instruct-BLIP, reproduced=true2025.03 | 47.2 | 6.9 | 27.5 | 2.2 | — | — | — | |
| M3IDBase Model=InstructBLIP-7B2026.04 | 47.2 | 6.9 | 27.5 | 2.2 | — | — | — | |
| AVISCBackbone=Instruct-BLIP, reproduced=true2025.03 | 46.7 | 6.7 | 28 | 2.6 | — | — | — | |
| AVISCBase Model=InstructBLIP-7B2026.04 | 46.7 | 6.7 | 28 | 2.6 | — | — | — | |
| CASBase Model=LLaVA-1.52026.05 | 46.7 | 4.4 | 17.2 | 1.6 | — | 0.1 | — | |
| VCDBackbone=LLaVA-1.5-7B2026.05 | 46.7 | 6.1 | 25.8 | 2.1 | — | — | — | |
| AVISCBackbone=LLaVA-v1.5-7B, reproduced=true2025.03 | 46.6 | 6.3 | 25.6 | 2 | — | — | — | |
| VCDBackbone=LLaVA-v1.5-7B, reproduced=true2025.03 | 46.5 | 6.7 | 27.8 | 2 | — | — | — | |
| Base (Comparable)Backbone=Instruct-BLIP2025.03 | 46.4 | 8.4 | 31.1 | 2.6 | — | — | — | |
| InstructBLIP-7BBase Model=InstructBLIP-7B2026.04 | 46.4 | 8.4 | 31.1 | 2.6 | — | — | — | |
| ICDBackbone=LLaVA-1.5-7B2026.05 | 46.3 | 6.3 | 25.8 | 2.2 | — | — | — | |
| EVASBackbone=LLaVA-1.5-7B2026.05 | 46.1 | 4.7 | 18.8 | 1.6 | — | — | — | |
| OPERABackbone=LLaVA-1.5-7B2026.05 | 45.6 | 4.6 | 18.9 | 1.6 | — | — | — | |
| PSRDBase Model=LLaVA-Next-7B2026.04 | 45.3 | 4.7 | 21.1 | 1.2 | — | — | — | |
| AVISCBackbone=LLaVA-1.5-7B2026.05 | 45.2 | 7.1 | 27.6 | 2.1 | — | — | — | |
| LLaVA-1.5-7BBackbone=LLaVA-1.5-7B2026.05 | 45 | 8.3 | 32 | 2.2 | — | — | — | |
| Base (Comparable)Backbone=LLaVA-v1.5-7B2025.03 | 44.5 | 8 | 31 | 2.2 | — | — | — | |
| SSLBase Model=LLaVA-1.52026.05 | 44.2 | 8.3 | 31.6 | 3.4 | — | 0.4 | — | |
| TAMEBackbone=LLaVA-1.5-7B2026.05 | 40.6 | 12 | 35.6 | 3.4 | — | — | — | |
| PAIBase Model=Shikra2026.05 | 33.5 | 8.2 | 21.4 | 1.9 | — | 1.4 | — | |
| SFT + DA-DPOBase Model=Qwen2.5-VL-7B2026.06 | — | — | — | — | — | — | 44.8 | |
| SFT + DA-DPOBase Model=Qwen2.5-VL-72B2026.06 | — | — | — | — | — | — | 49.8 | |
| SFT + DPOBase Model=Qwen2.5-VL-7B2026.06 | — | — | — | — | — | — | 42.5 | |
| SFT + DPOBase Model=Qwen2.5-VL-72B2026.06 | — | — | — | — | — | — | 46.8 | |
| SFT + HA-DPOBase Model=Qwen2.5-VL-72B2026.06 | — | — | — | — | — | — | 48.5 | |
| SFT + SimPOBase Model=Qwen2.5-VL-7B2026.06 | — | — | — | — | — | — | 42.8 | |
| SFT + SimPOBase Model=Qwen2.5-VL-72B2026.06 | — | — | — | — | — | — | 47 | |
| SFT + VIGILBase Model=Qwen2.5-VL-7B2026.06 | — | — | — | — | — | — | 46.5 | |
| SFT + VIGILBase Model=Qwen2.5-VL-72B2026.06 | — | — | — | — | — | — | 52.2 | |
| SFT onlyBase Model=Qwen2.5-VL-7B2026.06 | — | — | — | — | — | — | 41.2 | |
| SFT onlyBase Model=Qwen2.5-VL-72B2026.06 | — | — | — | — | — | — | 45.3 |