AI Paper Atlas

Research map / Vision-language and multimodal / Multimodal LLMs

ONLY: One-Layer Intervention Sufficiently Mitigates Hallucinations in Large Vision-Language Models

Zifu Wan, Ce Zhang, Silong Yong, Martin Q. Ma, Simon Stepputtis, Louis-Philippe Morency, Deva Ramanan, Katia Sycara, Yaqi Xie

ICCV 2025 · Vision-language and multimodal › Multimodal LLMs · approach: Decoding-time hallucination mitigation · 28 citations (OpenAlex)

Read the paper

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
apply ONLY one-layer textual-enhanced decoding on top of regular decodingLLaVA-1.5 · POPE MS-COCO Randomaccuracy: 83.13 -> 89.70Improved
apply ONLY decodingLLaVA-1.5 (max token 128) · CHAIRCHAIR_S: 55.0 -> 49.8Improved
apply ONLY decodingLLaVA-1.5 · MME-HallucinationMME total score: 562.50 -> 635.55Improved
apply ONLY decodingLLaVA-1.5 · MMVPaccuracy: 22.67% -> 28.00%Improved
use ONLY instead of two-query contrastive decoding (VCD, M3ID)LLaVA-1.5 · inference latencyrelative latency: x2.01 (VCD) -> x1.07Improved
select attention heads by text-to-visual entropy ratio instead of ratio of attention sums (ablation)ONLY textual enhancement · POPE and CHAIRPOPE accuracy: 84.20 -> 84.91; CHAIR_S 23.1 -> 20.0Improved
use entropy-ratio head selection instead of zeroing visual attention, adding noise, or doubling textual attention (ablation)ONLY textual enhancement · POPE accuracyaccuracy: 84.26 -> 84.91 (vs zeroing visual attention)Improved
add collaborative term (alpha_1 = 3 instead of 0) (ablation)ONLY adaptive decoding · POPE and CHAIR_Saccuracy: 88.13 -> 89.70; CHAIR_S 23.5 -> 20.0Improved
increase alpha_1 beyond 3 (to 4) (ablation)ONLY adaptive decoding · POPEaccuracy: 89.70 -> 88.37Hurt
add contrastive term (alpha_2 = 1 instead of 0) (ablation)ONLY adaptive decoding · POPE and CHAIR_Saccuracy: 86.50 -> 89.70; CHAIR_S 24.8 -> 20.0Improved
apply intervention at the initial layer instead of other layers (ablation)ONLY · POPEaccuracyNo clear effect
ONLY on InstructBLIP count subset MMEInstructBLIP · MME-Hallucination Count and Positionsubset score: Count 77.78 vs VCD 91.67; Position 74.44 vs Regular 79.58Mixed

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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