Object Hallucination on POPE (Random/Popular/Adversarial Metrics) benchmark leaderboard · SOTA2 Research

Object Hallucination on POPE (Random/Popular/Adversarial Metrics)

91.5F1 Score (Random)Qwen2.5-VL-3B + P2-DPO

Evaluation Results

Method	Links
Qwen2.5-VL-3B + P2-DPO2026.06		91.5	—	—	88.48	—	86.88	88.95
Qwen2.5-VL-3B + DPORLHF-V2026.06		90.4	—	—	88.59	—	86.41	88.46
LLaVA-1.5-7B + HA-DPO2026.06		90.25	—	—	87.81	—	82.54	86.86
RC-DPO2026.05		90.07	90.73	88.03	87.55	86	85.73	—
Qwen2.5-VL-7B + P2-DPO2026.06		90.05	—	—	87.79	—	86.79	88.21
LLaVA-1.5-7B + V-DPORLHF-V2026.06		89.9	—	—	87.91	—	84.05	87.28
LLaVA-1.5-7B + P2-DPO2026.06		89.87	—	—	87.91	—	84.53	87.44
Qwen2.5-VL-3B2026.06		89.79	—	—	88.05	—	86.26	88.03
Qwen2.5-VL-7B + DPORLHF-V2026.06		89.72	—	—	86.81	—	84.99	87.17
LLaVA-1.5-7B + V-DPOSAD2026.06		89.57	—	—	87.62	—	83.77	86.98
LLaVA-1.5-7B2026.06		89.12	—	—	84.36	—	81.8	85.1
LLaVA-1.5-7B + DPORLHF-V2026.06		89.12	—	—	85.94	—	84.03	86.38
Qwen2.5-VL-7B2026.06		89.01	—	—	86.78	—	84.92	86.9
OPA-DPO2026.05		88.09	89.23	87.4	86.33	86.13	85.16	—
SID2026.05		87.66	88.87	86.97	85.8	86.1	85.06	—
VCD2026.05		87.23	88.57	86.97	85.74	86.07	84.87	—
LLaVA-1.5-7B + VCD2026.06		87.16	—	—	85.06	—	81.33	84.51
Vanilla2026.05		86.98	88.37	86.73	85.41	85.73	84.48	—
RLAIF-V2026.05		86.34	87.9	86.73	85.22	85.4	83.97	—
RC-DPO2026.05		86.2	87.63	86	84.46	84.87	83.41	—
OPA-DPO2026.05		85.53	87.27	85.87	84.19	85.2	83.57	—
SID2026.05		84.86	86.77	85.73	83.9	84.53	82.63	—
VCD2026.05		84.38	86.4	85.47	83.51	84.37	82.41	—
Vanilla2026.05		83.52	85.77	84.73	82.52	84.2	82.02	—
RLAIF-V2026.05		81.78	84.53	83.63	80.86	82.93	80.2	—