arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2610.22356cs.CV

DK-VLM:将几何作为视觉语言模型空间推理的位置

DK-VLM: Geometry as Position for Spatial Reasoning in Vision-Language Models

Harbin Institute of Technology(哈尔滨工业大学);Yinwang Intelligent Technology(银王智能科技);Huawei(华为)

Han Su, Meng Tian, Tianyu Huang, Mingyu Gao, Lennox Wei, Dongxu Li, Jianhua Han, Hang Xu, Xiaohe Wu, Wangmeng Zuo

AI总结:

DK-VLM通过双核注意力将几何作为位置引入视觉语言模型,结合参考帧合成监督,在6个空间基准中取得最优平均得分,提升了多步空间推理能力并保留通用视频理解能力。

摘要:

视觉语言模型仍难以理解跨视角的3D空间关系,以及在观察者移动时更新空间判断。一种常见方法是将3D重建特征融合到视觉token中,但在空间监督有限的情况下,这些特征与预训练语义表示的对齐仍具挑战性。我们提出DK-VLM,其采用双核注意力(Dual-Kernel Attention),该注意力同时从视觉token的输入位置及其在场景中的几何位置计算注意力。语义核保留预训练模型的原生注意力和位置编码,而单独的几何核则利用估计的3D点和相机位姿,通过点位移和相对视角变换直接调节注意力。随后,一个学习到的门控融合两者的输出。因此,几何以位置的形式进入注意力,而非作为融合的重建特征。我们进一步引入参考帧合成(Reference-Frame Composition, RC)监督,以训练模型在连续判断中更新其参考帧,该监督提供从验证的单步注释中导出的逐步转向和航向目标。在对比方法中,DK-VLM-8B在6个空间基准中取得最高平均得分,并在VSI-Bench(74.33)和ReVSI-32(58.28)上取得最高总分。RC监督进一步提升多步空间推理能力,而完整模型保留了通用视频理解能力。项目页面:this https URL。

英文摘要:

Vision-language models still struggle to understand 3D spatial relations across views and update spatial judgments as the observer moves. A common approach fuses 3D reconstruction features into visual tokens, but aligning these features with pretrained semantic representations remains challenging under limited spatial supervision. We propose DK-VLM with Dual-Kernel Attention, which computes attention from both a visual token's input position and its geometric position in the scene. A semantic kernel retains the pretrained model's native attention and positional encoding, while a separate geometric kernel uses estimated 3D points and camera poses to directly modulate attention through point displacements and relative view transformations. Then, a learned gate fuses their outputs. Geometry thus enters attention as position rather than as fused reconstruction features. We further introduce Reference-Frame Composition (RC) supervision to train the model to update its reference frame across successive judgments. It provides stepwise turn and heading targets derived from verified single-step annotations. Among the compared methods, DK-VLM-8B achieves the highest average score across six spatial benchmarks and the highest overall scores on VSI-Bench (74.33) and ReVSI-32 (58.28). RC supervision further improves multi-step spatial reasoning, while the full model preserves general video understanding. Project page: https://sumuru789.github.io/DK-VLM-website/.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑