arXiv：cs.AI· Yike Wu, Keyan Chen, Qixiong Wang, Xi Xiang, Maofei Que, Xiaolong Jiang, Xiaoqin Zhang, Ling Shao, Shijian Lu·· 16 小时前
ReTAC：面向视频空间推理的残差式测试时智能体计算框架
ReTAC: Residual Test-Time Agentic Computation for Video Spatial Reasoning
AI 导读
ReTAC 是一个免训练的多智能体框架，用于视频空间推理，主张测试时计算应当"残差化"：只为基座模型未解决的需求分配额外算力，并在局部失败后仅重算受影响的状态。它通过规划步骤获取缺失的空间证据，并用阶段对齐验证将失败定位到规划、感知或推理环节，依赖关系恢复机制则复用未受影响的状态。在三个主流 MLLM 和两个基准上，ReTAC 取得了对比方法中的最高综合得分。
正文
Abstract:Test-time agentic computation can improve multimodal large language models (MLLMs) through planning, tool use, iterative perception, and reflection, but indiscriminate computation may duplicate resolved work, introduce noisy evidence, and perturb correct predictions. We propose that such computation should be residual: allocate additional computation only to requirements unresolved by the base model and, after a localized failure, recompute only the affected states. We instantiate this formulation in ReTAC, a training-free multiagent framework for video spatial reasoning. ReTAC organizes supplementary computation around unresolved requirements. It plans ordered steps to acquire missing spatial evidence and combines the observations with compatible native evidence for reasoning. In addition, it uses stage-aligned verification to localize failures to planning, perception, or reasoning. Dependency-based recovery recomputes only affected states while reusing unaffected ones. Extensive experiments show that ReTAC achieves the highest overall score among the compared methods across three mainstream MLLMs and two widely adopted benchmarks. Further analyses show that gains from additional computation depend not only on how much is used, but also on where it is allocated and which valid intermediate states are preserved.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.16315 [cs.AI] |
| (or arXiv:2610.16315v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.16315 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yike Wu [view email]
[v1]
Mon, 12 Oct 2026 07:39:58 UTC (1,132 KB)
来源：arXiv：cs.AI · arxiv.org