Fox Framework Reduces Hallucinations in Vision-Language Models

Liu Yu, Can Chen, Ping Kuang, Zhikun Feng, Fan Zhou, Gillian Dobbie· June 29, 2026 View original

Summary

This research introduces Fox, a training-free inference-time framework that reduces object hallucination in Large Vision-Language Models (LVLMs) by dismantling "pathological shortcuts." Fox diagnoses structural misalignment using visual attention entropy and applies targeted causal interventions to sever the shortcut path, achieving state-of-the-art faithfulness while preserving linguistic richness.

Large Vision-Language Models (LVLMs) are powerful but often suffer from "object hallucination," where they describe objects not present in an image. This issue is not merely due to attention intensity but stems from a deeper structural misalignment: specific attention heads can decouple from visual evidence, relying instead on language priors, creating a "pathological shortcut." Researchers have developed Fox (Faithfulness and Observational-flow via eXpression-rectification), a novel framework designed to dismantle these shortcuts during inference without requiring retraining. Fox first diagnoses structural misalignment by using a visual attention entropy probe to identify these "risky mediators" in an unsupervised manner. Once identified, Fox performs a targeted causal intervention by numerically saturating logits, effectively severing the shortcut path. A conflict-gated cooperative decoding strategy then reconciles the improved faithfulness with the model's natural fluency. Extensive experiments show Fox significantly outperforms existing methods, improving faithfulness by 29.1% while maintaining linguistic quality.

Why it matters

For AI developers and product managers working with LVLMs, Fox offers a crucial method to improve the reliability and trustworthiness of these models by directly addressing the problem of hallucination, making them more suitable for real-world applications.

How to implement this in your domain

  1. 1Integrate the Fox framework as an inference-time module for existing LVLM deployments.
  2. 2Apply the visual attention entropy probe to diagnose and localize hallucination-prone attention heads.
  3. 3Implement the targeted causal intervention via numerical logit saturation to sever pathological shortcuts.
  4. 4Utilize the conflict-gated cooperative decoding strategy to balance faithfulness and fluency.
  5. 5Evaluate the reduction in object hallucination and maintenance of linguistic richness in specific applications.

Who benefits

AI DevelopmentContent CreationE-commerceHealthcareEducation

Key takeaways

  • Object hallucination in LVLMs is caused by "pathological shortcuts" where attention heads decouple from visual evidence.
  • Fox is a training-free, inference-time framework to dismantle these shortcuts.
  • It uses visual attention entropy to diagnose misalignment and causal intervention to sever paths.
  • Fox significantly improves faithfulness in LVLMs while preserving linguistic quality.

Original post by Liu Yu, Can Chen, Ping Kuang, Zhikun Feng, Fan Zhou, Gillian Dobbie

"arXiv:2606.27596v1 Announce Type: cross Abstract: Large Vision-Language Models (LVLMs) exhibit sophisticated reasoning but remain susceptible to object hallucination. Deviating from the prevailing attention intensity assumption, we reveal a deeper dynamic structural misalignment:…"

View on X

Originally posted by Liu Yu, Can Chen, Ping Kuang, Zhikun Feng, Fan Zhou, Gillian Dobbie on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses