AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

Fuzzy Contrastive Decoding to Alleviate Object Hallucination in Large Vision-Language Models

Jieun Kim, Jinmyeong Kim, Yoonji Kim, Sung-Bae Cho

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Decoding-time hallucination mitigation · 1 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 FuzzyCD decodingLLaVA-1.5 · POPE MSCOCO Randomaccuracy (%): 82.90 -> 88.97Improved
add FuzzyCD decodingLLaVA-1.5 · POPE MSCOCO Popularaccuracy (%): 81.00 -> 86.10Improved
add FuzzyCD decodingLLaVA-1.5 · POPE MSCOCO Adversarialaccuracy (%): 78.60 -> 83.93Improved
add FuzzyCD decodingInstructBLIP · POPE MSCOCO Randomaccuracy (%): 80.63 -> 86.66Improved
add FuzzyCD decodingLLaVA-1.5 · ROPE multi-object homogeneousaccuracy (%): 31.88 -> 46.44Improved
add FuzzyCD decodingLLaVA-1.5 · ROPE multi-object wildaccuracy (%): 13.96 -> 21.12Improved
add FuzzyCD decodingLLaVA-1.5 · ROPE multi-object heterogeneousaccuracy (%): 3.98 -> 7.45Improved
add FuzzyCD decodingLLaVA-1.5 · MME total scorescore: 556.34 -> 641.66Improved
add FuzzyCD decoding versus VDD on colorLLaVA-1.5 · MME colorscore: 165 -> 160Hurt
add FuzzyCD decodingLLaVA-1.5 · MME object countscore: 106.67 -> 163.33Improved
scale model from 7B to larger VLMs (ablation)FuzzyCD · POPEaccuracy gain: 5.12%p LLaVA-1.5 -> 0.72%p QwenVL(32B)Mixed
use Sobel instead of median image filter (ablation)FuzzyCD · POPEaccuracy (%): 86.53 -> 89.33Improved

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