拆解嵌合体：视频幻觉的帧级风险与免训练缓解
Unstitching the Chimera: Frame-Level Risk and Train-Free Mitigation for Video Hallucination
提出帧级无参考风险指标 CH-Risk 和免训练两阶段干预 CH-M，可直接用于视频 MLLM 的幻觉检测与缓解，其注意力路由和词元校准思路可迁移到其他多模态生成任务。
原文依据 We introduce a frame-first study of Chimera Hallucination: model stitches visual segments that exist in space and time but do not belong to the same event chain
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论幻觉限制了多模态大语言模型（MLLMs）的可靠性，而在视频中尤为有害，因为错误表现为扭曲的叙事而非单帧错误。我们引入对嵌合体幻觉的帧优先研究：模型将存在于空间和时间中但不属于同一事件链的视觉片段拼接起来，产生虚假的连续故事。我们提出 CH-Risk，一种针对该失败模式的单次前向、无参考风险估计。CH-Risk 结合两个互补信号：SegCoverage@α（SCR@α）衡量覆盖大部分文本到帧支持所需的事件片段数量，暴露长程拼接；早期时序通路对齐（AETP）衡量支持与早中层形成的时序通路之间的秩一致性，暴露阶段不匹配。为了将风险转化为校正，我们进一步提出 CH-M(itigation)，一种免训练的两阶段干预。片段对齐的阶段对齐帧路由（sSAFR）在中层 softmax 之前重新加权帧，将注意力导向一小部分通路对齐的片段。残差词元校准（RTC）随后稳定所选片段内的词元使用。在 9 个基准和 6 个 VideoLLM 上的大量实验表明，CH-Risk 能够预测嵌合体幻觉，且 CH-M 持续减少幻觉并提高任务准确率，开销可忽略（延迟低于 5%，内存低于 2.5%，FLOPs 约 1%）。
英文原文摘要 1481 字符
Hallucination limits the reliability of multimodal large language models (MLLMs), and it is particularly damaging in video where errors manifest as distorted narratives rather than single-frame mistakes. We introduce a frame-first study of **Chimera Hallucination**: model stitches visual segments that exist in space and time but do not belong to the same event chain, producing a spurious continuous story. We introduce **CH-Risk**, a single-forward, reference-free risk estimate tailored to this failure mode. CH-Risk combines two complementary signals: SegCoverage@\alpha (\mathrm SCR @\alpha) measures how many event segments are needed to cover most text-to-frame support, exposing long-range stitching; Alignment with Early Temporal Pathway (AETP) measures rank consistency between support and the temporal pathway formed in early-middle layers, exposing stage mismatch. To turn risk into correction, we further propose **CH-M(itigation)**, a train-free two-stage intervention. Segment-aligned Stage-Aligned Frame Routing (sSAFR) re-weights frames before the mid-layer softmax to route attention toward a small set of pathway-aligned segments. Residual Token Calibration (RTC) then stabilizes token usage within selected segments. Extensive experiments across 9 benchmarks and 6 VideoLLMs show that CH-Risk can predict Chimera and that CH-M consistently reduces hallucination and improves task accuracy with negligible overhead (sub-5% latency, sub-2.5% memory, \~1% FLOPs).