# STEAR: Layer-Aware Spatiotemporal Evidence Intervention for Hallucination Mitigation in Video Large Language Models

Source: [STEAR: Layer-Aware Spatiotemporal Evidence Intervention for Hallucination Mitigation in Video Large Language Models](<https://www.sota2.com/research/paper/69d33def7d0172140ec735ff>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-07T05:00:04.608Z

Authors: Linfeng Fan, Yuan Tian, Ziwei Li, Zhiwu Lu

Published: 2026-04-03T13:52:57.000Z

## Abstract

Video Large Language Models (Video-LLMs) remain prone to spatiotemporal hallucinations, often generating visually unsupported details or incorrect temporal relations. Existing mitigation methods typically treat hallucination as a uniform decoding failure, applying globally shared correction rules. We instead observe that decoder layers contribute differently to visual grounding and later linguistic composition, indicating that intervention must be layer-aware. Based on this insight, we propose STEAR, a layer-aware spatiotemporal evidence intervention framework. STEAR identifies high-risk decoding steps and selects token-conditioned visual evidence from grounding-sensitive middle layers. It uses this shared evidence for two coupled purposes: restoring missing local grounding in middle layers, and constructing temporally perturbed patch-level counterfactuals to falsify inconsistent reasoning during late-layer decoding. Consequently, STEAR mitigates both spatial and temporal hallucinations within an efficient single-encode inference framework. Experiments across representative Video-LLM backbones and challenging benchmarks demonstrate that STEAR consistently reduces hallucinations while improving faithfulness, temporal consistency, and robustness. Our results confirm that reliable video decoding relies on intervening on precise evidence at the right layer, rather than enforcing a global penalty. The code is provided in the Supplementary Material.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Video Understanding on VidHalluc](<https://www.sota2.com/research/sota/video-understanding-on-vidhalluc.md>): Video Understanding; VidHalluc; Accuracy: 83.84.

- [Video Understanding on EventHallusion](<https://www.sota2.com/research/sota/video-understanding-on-eventhallusion.md>): Video Understanding; EventHallusion; Accuracy: 72.37.

- [Video Understanding on MVBench](<https://www.sota2.com/research/sota/video-understanding-on-mvbench.md>): Video Understanding; MVBench; Accuracy: 68.21.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2604.03045>)

[PDF](<https://arxiv.org/pdf/2604.03045.pdf>)