Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

YARD: Y-Architecture Register Decoding for Efficient Hallucination Mitigation in Large Vision-Language Models

About

Contrastive decoding (CD) seeks to mitigate hallucinations in Large Vision-Language Models (LVLMs) by contrasting the output distributions of a standard model and a visually degraded model. However, existing training-free CD methods suffer from sub-optimal degraded branches: completely dropping visual tokens is too extreme and induces language hallucinations, while corrupting input images offers coarse control over visual evidence and suffers from high inference latency due to requiring two full forward passes. To address these dilemmas, we propose YARD, a training-free Y-Architecture Register Decoding framework. Motivated by the observation that reliable text-to-vision grounding predominantly emerges in the middle decoder layers, YARD constructs the degraded branch internally by sharing shallow-layer computations and branching exactly at this critical stage. For the degraded branch, YARD replaces patch-level visual tokens with register tokens, which preserve global image semantics but lack fine-grained local evidence. This image-aware yet locally under-grounded design provides a faithful contrastive signal without extreme modality mismatch, while the Y-architecture strictly avoids a costly second forward pass. Extensive experiments on generative and discriminative hallucination benchmarks demonstrate that YARD consistently achieves state-of-the-art hallucination mitigation across multiple LVLMs, alongside a significant reduction in inference latency.

Ting Chen, Geng Li, Guohao Chen, Yu Hu, Guan Huang, Mai Chen, Langsheng Lei, Jun Du• 2026

Related benchmarks

TaskDatasetResultRank
Generative HallucinationAMBER Generative
Coverage (%)48.9
81
Object Hallucination EvaluationMSCOCO POPE--
71
Hallucination EvaluationMME Hallucination
Existence Score185
61
Discriminative Object HallucinationPOPE MSCOCO Adversarial
Accuracy87.33
43
Generative HallucinationObject-HalBench
CHAIR_S Score19.7
43
Discriminative Object HallucinationPOPE MSCOCO (Random)
Accuracy90.77
29
Discriminative Object HallucinationPOPE MSCOCO Popular
Accuracy89.07
29
Discriminative Hallucination EvaluationAMBER-d
F1 Score84.1
23
Showing 8 of 8 rows

Other info

Follow for update