ReLoop: "Seeing Twice and Thinking Backwards" via Closed-loop Training to Mitigate Hallucinations in Multimodal understanding

TL;DR

ReLoop introduces a closed-loop training framework for multimodal models that enhances their internal consistency and significantly reduces hallucinations by integrating multiple feedback mechanisms during training.

Contribution

The paper presents ReLoop, a novel training method that enforces multimodal consistency through a ring-shaped structure with integrated feedback modules, addressing hallucinations internally.

Findings

ReLoop reduces hallucination rates across multiple benchmarks.

The framework improves semantic and visual consistency in MLLMs.

ReLoop enhances interpretability through attention supervision.

Abstract

While Multimodal Large Language Models (MLLMs) have achieved remarkable progress in open-ended visual question answering, they remain vulnerable to hallucinations. These are outputs that contradict or misrepresent input semantics, posing a critical challenge to the reliability and factual consistency. Existing methods often rely on external verification or post-hoc correction, lacking an internal mechanism to validate outputs directly during training. To bridge this gap, we propose ReLoop, a unified closed-loop training framework that encourages multimodal consistency for cross-modal understanding in MLLMs. ReLoop adopts a ring-shaped structure that integrates three complementary consistency feedback mechanisms, obliging MLLMs to "seeing twice and thinking backwards". Specifically, ReLoop employs the frozen Consistency Feedback Plugin (CFP), comprising semantic reconstruction, visual description, and an attention supervision module for attention alignment. These components collectively enforce semantic reversibility, visual consistency, and interpretable attention, enabling the model to correct its outputs during training. Extensive evaluations and analyses demonstrate the effectiveness of ReLoop in reducing hallucination rates across multiple benchmarks, establishing a robust method for hallucination mitigation in MLLMs. We will release our source code and data in the camera-ready version.