Mitigating Hallucinations in Video Large Language Models via Spatiotemporal-Semantic Contrastive Decoding

TL;DR

This paper introduces a novel decoding method for Video Large Language Models that reduces hallucinations by contrasting disrupted spatiotemporal and semantic features, improving factual accuracy without sacrificing understanding.

Contribution

The paper presents Spatiotemporal-Semantic Contrastive Decoding, a new approach that explicitly targets hallucination root causes by disrupting and contrasting video features during decoding.

Findings

Significantly reduces hallucinations in video LLM outputs

Maintains core video understanding and reasoning capabilities

Demonstrates robustness across complex scenarios

Abstract

Although Video Large Language Models perform remarkably well across tasks such as video understanding, question answering, and reasoning, they still suffer from the problem of hallucination, which refers to generating outputs that are inconsistent with explicit video content or factual evidence. However, existing decoding methods for mitigating video hallucinations, while considering the spatiotemporal characteristics of videos, mostly rely on heuristic designs. As a result, they fail to precisely capture the root causes of hallucinations and their fine-grained temporal and semantic correlations, leading to limited robustness and generalization in complex scenarios. To more effectively mitigate video hallucinations, we propose a novel decoding strategy termed Spatiotemporal-Semantic Contrastive Decoding. This strategy constructs negative features by deliberately disrupting the spatiotemporal consistency and semantic associations of video features, and suppresses video hallucinations through contrastive decoding against the original video features during inference. Extensive experiments demonstrate that our method not only effectively mitigates the occurrence of hallucinations, but also preserves the general video understanding and reasoning capabilities of the model.