OmniGround：面向真实复杂场景的综合时空视觉定位基准
OmniGround: A Comprehensive Spatio-Temporal Grounding Benchmark for Real-World Complex Scenarios
STVG 被拆成「高层时序定位 + 细粒度时空传播」两阶段且免训练，可直接套用到现有多模态大模型上做视频目标定位，无需微调即可提升小目标与遮挡场景的 m_tIoU/m_vIoU。
原文依据 we propose PG-TAF, a training-free two-stage framework decomposing STVG into high-level temporal grounding and fine-grained spatio-temporal propagation
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论时空视频定位（STVG）旨在根据自然语言描述在视频中定位目标物体。尽管多模态大语言模型已展现出潜力，但当前模型与涉及多样物体和复杂查询的真实世界需求之间仍存在显著差距。我们将此归因于基准覆盖范围有限，导致模型表现出类别偏置、推理过度简化以及语言鲁棒性差。为解决这些局限，我们提出 OmniGround，一个涵盖 81 个类别、包含 3,475 段视频及复杂真实世界查询的综合基准。我们提出前向-后向-精修（FBR）标注流程以获得高质量标签，并提出 DeepSTG，一个超越表面统计、可量化数据集质量的系统化评测框架。评测显示，在复杂真实世界场景中性能平均下降 10.4%，在小目标/遮挡目标以及复杂空间关系上尤为明显。受此启发，我们提出 PG-TAF，一个免训练的两阶段框架，将 STVG 分解为高层时序定位与细粒度时空传播。实验表明，PG-TAF 在 OmniGround 上的 m_tIoU 与 m_vIoU 分别提升 25.6% 和 35.6%，并在四个基准上取得一致增益。
英文原文摘要 1276 字符
Spatio-Temporal Video Grounding (STVG) aims to localize target objects in videos based on natural language descriptions. While Multimodal Large Language Models have shown promise, a significant gap remains between current models and real-world demands involving diverse objects and complex queries. We attribute this to limited benchmark scope, causing models to exhibit category bias, oversimplified reasoning, and poor linguistic robustness.To address these limitations, we introduce OmniGround, a comprehensive benchmark with 3,475 videos spanning 81 categories and complex real-world queries. We propose the Forward-Backward-Refinement (FBR) annotation pipeline for high-quality labels and DeepSTG, a systematic evaluation framework quantifying dataset quality beyond superficial statistics. Evaluations reveal performance average drops of 10.4% on complex real-world scenes, particularly with small/occluded objects and intricate spatial relations. Motivated by these, we propose PG-TAF, a training-free two-stage framework decomposing STVG into high-level temporal grounding and fine-grained spatio-temporal propagation. Experiments demonstrate PG-TAF achieves 25.6% and 35.6% improvements in m_tIoU and m_vIoU on OmniGround with consistent gains across four benchmarks.