AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Context Blindness in DPO: Mitigating Object Hallucination in MLLMs via Context-Calibrated Preference Optimization

Byungoh Ko, Jinyoung Park, Jongha Kim, Jeehye Na, Jaewon Cho, Hyunwoo Kim

ECCV 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Preference tuning against hallucination · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add context-calibrated loss (C2-DPO) over C-DPOQwen2-VL-Instruct-2B · Object HalBench responsehallucination rate: 2.5 -> 1.6Improved
add context-calibrated loss over C-DPOLLaVA-v1.5-7B · Object HalBench responsehallucination rate: 5.9 -> 4.8Improved
add context-calibrated loss over C-DPOLLaVA-v1.5-7B · AMBERHallucination Score: 14.9 -> 13.8Improved
replace C-DPO with C2-DPOQwen2-VL-Instruct-2B · MM-Vetoverall: 48.2 -> 49.1 (base 49.9)Mixed
apply C2-DPOLLaVA-v1.5-7B · ScienceQA / TextVQAaccuracy: TextVQA 58.2 -> 58.2No clear effect
add L_c and DPO(x') jointly to DPO(x) (ablation)C2-DPO · Object HalBench response (LLaVA)hallucination rate: 5.9 -> 4.8Improved
add DPO on degraded context only (ablation)C2-DPO · Object HalBench response (LLaVA)hallucination rate: 5.9 -> 7.1Hurt
add L_c only (no DPO(x')) (ablation)C2-DPO · Object HalBench response (Qwen)hallucination rate: 2.5 -> 3.5Hurt
add calibration to SimPO (ablation)SimPO · Object HalBench responsehallucination rate: 3.9 -> 3.3Improved
add calibration to RDPO (ablation)RDPO · AMBERHallucination Score: 38.7 -> 29.0Improved
add calibration to DPO in text-only setting (ablation)DPO · AlpacaEval 2 LCwin rate: 24.1 -> 25.9Improved
degrade by removing caption rather than image (ablation)C2-DPO · AMBERHallucination Score: 18.2 -> 16.1Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy