STORM: Internalized Modeling for Spatial-Temporal Reasoning in Video-Language Models

• URL: http://arxiv.org/abs/2605.26014v1
• Date: Mon, 25 May 2026 16:33:00 GMT
• Title: STORM: Internalized Modeling for Spatial-Temporal Reasoning in Video-Language Models
• Abstract summary: Video reasoning tasks require tracking motion, temporal order, and evolving visual states across frames.<n>Existing methods built on vision-language models (LVLMs) often address this challenge by externalizing reasoning through chain-of-thought (CoT)<n>We propose STORMS, a two-stage framework that teaches LVLMs to reason through bounded continuous latent trajectories instead of explicit textual CoT.
• Score: 81.32710031596591
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Many video reasoning tasks require tracking motion, temporal order, and evolving visual states across frames. Existing methods built on large vision-language models (LVLMs) often address this challenge by externalizing reasoning through textual chain-of-thought (CoT), keyframe selection, repeated frame reinsertion, or external tool use. While effective, such pipelines increase inference-time latency and engineering complexity, and they force temporal-visual evidence to be serialized into text or repeatedly re-encoded from frames. Inspired by the intuition that visual reasoning can occur implicitly before verbalization, we propose STORMS (Spatial-Temporal reasOning via inteRnalized Modeling), a two-stage framework that teaches LVLMs to reason through bounded continuous latent trajectories instead of explicit textual CoT. In Stage I, STORMS aligns latent tokens with thought-video representations derived from generated videos, grounding the latent states in dynamic visual evidence. In Stage II, the model is further trained with answer-only supervision, encouraging the reasoning process to be internalized without step-by-step annotations. Generated thought videos are used only during training; at inference, STORMS performs a bounded latent rollout without regenerating videos, reinserting frames, or invoking external visual tools. Experiments on VideoMME, MVBench, TempCompass, and MMVU show that STORMS improves video reasoning accuracy while substantially reducing inference overhead compared with tool or video-generation-based reasoning pipelines.

Related papers

• TimeThink: Reasoning with Time for Video LLMs [28.549802513886046]
  We propose TimeThink, a reinforcement learning framework that guides temporal evidence discovery in Video-LLMs.<n>Our key idea is to treat temporal clue steps as the fundamental optimization primitive of video reasoning.<n>To enable scalable training, we construct TimeThink-RFT-20K, a dataset with automatically derived temporal evidence segments.
  arXiv  Detail & Related papers  (2026-07-06T13:50:15Z)
• InternVideo3: Agentify Foundation Models with Multimodal Contextual Reasoning [61.87294123934038]
  We present InternVideo3, a framework enhancing multimodal understanding in video tasks.<n>MCR treats understanding as a closed-loop process over a shared, evolving context.<n>We show InternVideo3 achieves strong performance on benchmarks like Video-MME, MLVU, and Ego.
  arXiv  Detail & Related papers  (2026-06-10T15:17:08Z)
• Act2See: Emergent Active Visual Perception for Video Reasoning [70.52373240573856]
  Vision-Language Models (VLMs) typically rely on static initial frames for video reasoning.<n>We introduce Act-to-See (Act2See), a novel framework that enables active visual perception.
  arXiv  Detail & Related papers  (2026-05-03T00:52:51Z)
• When Thinking Hurts: Mitigating Visual Forgetting in Video Reasoning via Frame Repetition [22.037040360505742]
  In Video Question Answering, models increasingly rely on self-generated text, sidelining visual inputs and causing hallucinations.<n>We propose FrameRepeat, an automated enhancement framework which features a lightweight repeat scoring module.<n>We show that FrameRepeat is both effective and generalizable in strengthening important visual cues during the reasoning process.
  arXiv  Detail & Related papers  (2026-03-17T08:41:54Z)
• TV-RAG: A Temporal-aware and Semantic Entropy-Weighted Framework for Long Video Retrieval and Understanding [14.570869250170139]
  TV-RAG is a training-free architecture that couples temporal alignment with entropy-guided semantics to improve long-video reasoning.<n>By weaving these temporal and semantic signals together, TV-RAG realises a dual-level reasoning routine that can be grafted onto any LVLM without re-training or fine-tuning.
  arXiv  Detail & Related papers  (2025-12-29T14:10:22Z)
• FrameMind: Frame-Interleaved Video Reasoning via Reinforcement Learning [65.42201665046505]
  Current video understanding models rely on fixed frame sampling strategies, processing predetermined visual inputs regardless of the specific reasoning requirements of each question.<n>This static approach limits their ability to adaptively gather visual evidence, leading to suboptimal performance on tasks that require broad temporal coverage or fine-grained spatial detail.<n>We introduce FrameMind, an end-to-end framework trained with reinforcement learning that enables models to dynamically request visual information during reasoning through Frame-Interleaved Chain-of-Thought (FiCOT)<n>Unlike traditional approaches, FrameMind operates in multiple turns where the model alternates between textual reasoning and active visual perception, using tools to extract
  arXiv  Detail & Related papers  (2025-09-28T17:59:43Z)
• Video-LLMs with Temporal Visual Screening [59.18455762289321]
  Temporal Visual Screening (TVS) is a new task that universally pre-processes video question answering and instruction tuning data.<n>TVS is formulated as a modular front-end adapter task that can be seamlessly integrated into both Video Instruction Tuning (training) and Video Question Answering (inference) pipelines.<n> Experiments demonstrate that incorporating TVS yields relative gains of 7.33% (training) and 34.6% (inference)
  arXiv  Detail & Related papers  (2025-08-27T14:33:32Z)
• Episodic Memory Representation for Long-form Video Understanding [52.33907540905242]
  Large Video Language Models excel at general video understanding but struggle with long-form context window limits.<n>We introduce Video-EM, a training free framework inspired by the principles of human memory.<n>Video-EM achieves performance gains of 4-9 percent over respective baselines while utilizing fewer frames.
  arXiv  Detail & Related papers  (2025-08-13T04:33:07Z)
• Emergent Temporal Correspondences from Video Diffusion Transformers [30.83001895223298]
  We introduce DiffTrack, the first quantitative analysis framework designed to answer this question.<n>Our analysis reveals that query-key similarities in specific, but not all, layers play a critical role in temporal matching.<n>We extend our findings to motion-enhanced video generation with a novel guidance method that improves temporal consistency of generated videos without additional training.
  arXiv  Detail & Related papers  (2025-06-20T17:59:55Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.