CoFi-Dec: Hallucination-Resistant Decoding via Coarse-to-Fine Generative Feedback in Large Vision-Language Models

2025-12-29

AI Summary

Large Vision-Language Models (LVLMs) frequently generate hallucinated content inconsistent with input images, hindering real-world deployment. To address this, we propose CoFi-Dec—a training-free, model-agnostic decoding framework. It employs a coarse-to-fine visual-conditioned generative self-feedback mechanism to construct hierarchical visual hypotheses, and introduces a Wasserstein-distance-driven geometric consistency constraint to align decoding trajectories across multiple hypothesis scales. The method is plug-and-play, requiring no fine-tuning or additional parameters. Evaluated on six mainstream hallucination benchmarks, CoFi-Dec significantly reduces both entity-level and semantic-level hallucinations, consistently outperforming existing decoding strategies across diverse LVLMs. Its core innovations lie in the synergistic integration of generative self-feedback, multi-scale visual hypothesis modeling, and Wasserstein-based distributional alignment—enabling robust, vision-grounded text generation without architectural or training modifications.

Technology Category

• Computer Vision: Large Vision Models
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large Vision-Language Models (LVLMs) have achieved impressive progress in multi-modal understanding and generation. However, they still tend to produce hallucinated content that is inconsistent with the visual input, which limits their reliability in real-world applications. We propose extbf{CoFi-Dec}, a training-free decoding framework that mitigates hallucinations by integrating generative self-feedback with coarse-to-fine visual conditioning. Inspired by the human visual process from global scene perception to detailed inspection, CoFi-Dec first generates two intermediate textual responses conditioned on coarse- and fine-grained views of the original image. These responses are then transformed into synthetic images using a text-to-image model, forming multi-level visual hypotheses that enrich grounding cues. To unify the predictions from these multiple visual conditions, we introduce a Wasserstein-based fusion mechanism that aligns their predictive distributions into a geometrically consistent decoding trajectory. This principled fusion reconciles high-level semantic consistency with fine-grained visual grounding, leading to more robust and faithful outputs. Extensive experiments on six hallucination-focused benchmarks show that CoFi-Dec substantially reduces both entity-level and semantic-level hallucinations, outperforming existing decoding strategies. The framework is model-agnostic, requires no additional training, and can be seamlessly applied to a wide range of LVLMs. The implementation is available at https://github.com/AI-Researcher-Team/CoFi-Dec.

Problem

• Reduces hallucinations in large vision-language models' outputs
• Aligns text predictions with multi-level visual grounding cues
• Improves reliability without requiring additional model training

Innovation

• Coarse-to-fine generative feedback for hallucination reduction
• Wasserstein-based fusion aligns multi-level visual predictions
• Training-free, model-agnostic decoding framework for LVLMs