AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Mitigating Object Hallucinations via Sentence-Level Early Intervention

Shangpin Peng, Senqiao Yang, Li Jiang, Zhuotao Tian

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Preference tuning against hallucination · 22 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
use context-aware DPO instead of standard DPO (ablation)SENTINEL · Object HalBenchCHAIR response-level: 10.1 -> 4.3Improved
use non-hallucinated context instead of hallucinated context (ablation)SENTINEL · Object HalBenchCHAIR response-level: 14.3 -> 4.3Improved
use non-hallucinated context instead of natural context (ablation)SENTINEL · Object HalBenchCHAIR response-level: 8.6 -> 4.3Improved
add iterative contextual bootstrapping (ICB) (ablation)SENTINEL · Object HalBenchCHAIR response-level: 5.3 -> 4.3Improved
add ICB (ablation)SENTINEL · MM-Vetoverall: 31.8 -> 32.6Improved
use original in-domain pairs instead of rewritten pairs (ablation)SENTINEL · AMBERaccuracy: 75.0 -> 76.1Improved
ignore uncertain objects instead of treating as factual (ablation)SENTINEL · Object HalBenchCHAIR response-level: 10.3 -> 4.3Improved
ignore uncertain objects instead of treating as hallucinated (ablation)SENTINEL · Object HalBenchCHAIR response-level: 8.3 -> 4.3Improved
use two detectors (Grounding DINO + YOLO World) instead of one (ablation)SENTINEL detector · Object HalBenchCHAIR response-level: 14.3 -> 6.6Improved
use only positive-free mixed data (50% negative-positive) (ablation)SENTINEL · Object HalBenchCHAIR response-level: 4.3 -> 4.8Hurt
add SENTINEL data to HA-DPOHA-DPO · Object HalBenchCHAIR response-level: 37.0 -> 8.0Improved
SENTINEL versus base model on general VQALLaVA-v1.5-7B · TextVQAaccuracy: 58.2 -> 58.2No clear effect

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