Search papers, labs, and topics across Lattice.
This paper introduces Adaptive Context inTegration (ACT), a training-free inference method to mitigate hallucinations in Large Vision-Language Models (LVLMs). ACT adaptively integrates contextual information by using spatio-temporal profiling to amplify visual exploration attention heads and marginalizing semantic queries to aggregate visual evidence. Experiments across diverse LVLMs show that ACT significantly reduces hallucinations and achieves competitive results on discriminative and generative benchmarks without compromising generation capabilities.
LVLM hallucinations can be slashed without retraining, simply by adaptively integrating contextual information during inference.
Large Vision-Language Models (LVLMs) frequently suffer from severe hallucination issues. Existing mitigation strategies predominantly rely on isolated, single-step states to enhance visual focus or suppress strong linguistic priors. However, these static approaches neglect dynamic context changes across the generation process and struggles to correct inherited information loss. To address this limitation, we propose Adaptive Context inTegration (ACT), a training-free inference intervention method that mitigates hallucination through the adaptive integration of contextual information. Specifically, we first propose visual context exploration, which leverages spatio-temporal profiling to adaptively amplify attention heads responsible for visual exploration. To further facilitate vision-language alignment, we propose semantic context aggregation that marginalizes potential semantic queries to effectively aggregate visual evidence, thereby resolving the information loss caused by the discrete nature of token prediction. Extensive experiments across diverse LVLMs demonstrate that ACT significantly reduces hallucinations and achieves competitive results on both discriminative and generative benchmarks, acting as a robust and highly adaptable solution without compromising fundamental generation capabilities.
