May 2026· arXiv.org· Vol abs/2605.06185· 0 citations· 57 references
Computer Science
TL;DR
This work proposes Event-Causal RAG, a lightweight retrieval-augmented framework for ultra-long and streaming video reasoning and introduces ECV-1H, an hour-scale long-video QA benchmark dedicated to directed event-causal reasoning, with all source videos exceeding one hour.
Abstract
Large vision-language models perform well on short- and medium-length video understanding but still struggle to maintain coherent event memory and recover long-range relationships in ultra-long videos. End-to-end methods are limited by visual-token growth and context length, while fixed-segment retrieval often fragments complete events and weakens state-transition modeling.We propose Event-Causal RAG (EC-RAG), a lightweight retrieval-augmented framework for ultra-long and streaming video reasoning. A dual visual-audio sentinel mechanism segments video streams into semantically complete events, represented as State-Event-State (SES) structures that organize observable pre-event states, central events, and post-event states as event-local causal transitions. These transitions are stored in dual vector-graph memory and temporally connected through entity-consistent trajectories. During question answering, bidirectional graph retrieval recovers relevant predecessor and successor events, and answers are generated using both structured memory and the corresponding video evidence.We further introduce ECV-1H, an hour-scale long-video QA benchmark dedicated to directed event-causal reasoning, with all source videos exceeding one hour. It covers over 150 hours of untrimmed video and contains 1,251 fully human-annotated QA pairs. EC-RAG improves overall accuracy by 4.96\%--11.67\% across three open-source video foundation models and achieves consistent gains across public datasets. On a single RTX 5090 GPU with 32 GB of memory, EC-RAG can continuously process videos while maintaining controlled streaming memory usage.
Reasoning temporal localization (RTL) requires a model to generate an answer that itself contains the time interval supporting it, coupling high-level reasoning with temporal grounding in a single response. To tackle this challenge, we propose the first event-centric video chain-of-thought framework, dubbed EventCoT. EventCoT first performs event-centric tokenization, converting the video into compact event tokens that enable efficient identification of question-relevant events. It then reasons within these events to generate the answer, grounding the time interval via embedding matching that aligns placeholder tokens with visual embeddings. EventCoT achieves state-of-the-art results on ActivityNet-RTL while using substantially fewer visual tokens than previous work, and attains strong zero-shot results on the grounded video question answering benchmark ReXTime. Our code will be released for research purposes.
Youngkil Song, Yoonjae Baek, Dongwon Kim et al.· 0 citations
This work forms an episodic multi-key representation that enables precise retrieval of fine-grained memories through a simple key-matching mechanism, and introduces a neighbor filtering mechanism to capture broader semantic context without the massive computational overhead of global memory construction.
Yeeun Choi, Youngbeom Yoo, Joon-Young Lee et al.· 0 citations
Long-horizon event-based action understanding remains underexplored because existing datasets largely comprise short, trimmed clips, while collecting native event streams with dense temporal annotations is costly. We introduce Event ActivityNet, a large-scale simulated-event benchmark derived from human-annotated, untrimmed ActivityNet videos. It comprises 3,263 videos, 200 action classes, and 106.94 hours, with matched 5-bin and 9-bin event-voxel representations, temporal action annotations, and timestamped captions. The benchmark supports annotated-segment action recognition, auxiliary event-language alignment, and causal online temporal action localization. We generate event voxels directly from non-interpolated source videos in decoded frame order, retain per-video rational nominal or average frame-rate metadata for approximate time mapping, and use action-center reconstruction LPIPS as a soft diagnostic of retained reconstructable content. We establish baselines for adaptive event framing, prompt-caption alignment, and event-only, RGB-only, and RGB-event localization. Under a progressive nested-scale training protocol, recognition Top-1 accuracy increases from 52.25 to 66.42, while online temporal localization average mAP improves from 21.7 to 29.0. Moreover, staged Event ActivityNet pretraining followed by native-event fine-tuning consistently outperforms target-only and joint-from-scratch training across multiple supervision budgets. Event ActivityNet provides a scalable benchmark for long-horizon event modeling, although native-camera evaluation remains essential for deployment-oriented conclusions.
Cheng-Yao Hong, Ting-Wei Lin, Yun-Chung Lai et al.· 0 citations
Recently, retrieval-augmented and memory-augmented methods have emerged as two promising paradigms for long-video question answering. However, existing methods typically rely on rigid, fixed-length temporal chunking (e.g., 10s) and static offline memory banks, which not only fragment coherent continuous events but also fail to adapt during real-time reasoning. Moreover, whether using multi-scale summaries or multimodal knowledge graphs, current approaches prioritize retrieval relevance while overlooking evidence sufficiency, often stopping to answer once only semantically relevant clues are retrieved, even when key temporal, causal, or fine-grained action evidence is still missing. To tackle these challenges, we propose REVEAL, a rubric-guided agent framework. As a foundation, we introduce an adaptive visual-similarity-based preprocessing pipeline that groups visually coherent adjacent frames into natural event units to construct an offline-online video memory---capturing global video context offline while dynamically maintaining question-conditioned memory online. Built upon this structured memory, REVEAL uses an automatically constructed rubric library to explicitly verify whether retrieved evidence satisfies sufficiency criteria, pinpoints missing clues upon verification failure, and directs targeted re-retrieval for complementary information. Without any extra training, REVEAL consistently outperforms both closed-source and open-source state-of-the-art methods across extensive experiments. These results show that explicitly verifying evidence sufficiency, rather than stopping at semantic relevance, retrieves the decisive clues that prior methods miss and yields more reliable long-video reasoning.
C.J. Yan, Yang Zhou, Meixing Shi et al.· 0 citations
Long-video understanding remains challenging for Multimodal Large Language Models (MLLMs) due to limited context length. Uniform sampling may miss crucial moments, while agent-based frame video understanding methods often evaluate frames independently, overlooking the temporal organization of videos. Ideally, evidence selection should mimic how humans answer questions about long videos: first locating the relevant segment from the global context, then zooming into local objects and details. We propose Temporal Tree of Thought T^3, a training-free framework for adaptive coarse-to-fine long-video understanding. T^3 constructs a question-agnostic hierarchical temporal tree via recursive temporally constrained clustering, where each node represents a contiguous segment with an informative key frame. During inference, T^3 performs an answer-retrieve-explore loop: it reasons over coarse representative frames, generates a search statement when evidence is insufficient, and expands relevant branches for finer-grained evidence. This process adaptively shifts the search target from temporal regions to specific objects and visual details to help video understanding. Experiments on VideoMME, LongVideoBench, and LVBench show that T^3 improves Qwen2.5-VL-7B by 0.5%, 4.6%, and 4.4%, respectively, under the same frame budget, demonstrating the effectiveness of structured temporal reasoning.
Grounded long-video question answering (Grounded LVQA) requires answering a question about a long video while localizing the short evidence interval that supports the answer. Recent agentic methods frame this task as multi-turn exploration with a single crop_video(start, end) action, which supports coarse-to-fine narrowing but provides no primitive for fine-to-coarse backtracking. As a result, these agents typically converge prematurely and cannot recover from an early mistake. We propose VideoTreeSearch (VTS), a framework that casts grounded LVQA as iterative self-correcting search over an adaptive temporal tree. VTS constructs a non-uniform tree from visual scene boundaries so that each node corresponds to a semantically coherent segment, and trains an agent to navigate the tree through four discrete operations: zoom_in, zoom_out, shift, and answer. These operations expose backtracking and recovery as explicit, learnable primitives rather than implicit behaviors. To train this navigation, we introduce a trajectory synthesis pipeline that produces multi-step paths through the tree, including deliberate detours into incorrect branches followed by recovery. We use these trajectories for supervised fine-tuning, followed by reinforcement learning with grounding and answer-accuracy rewards. On three Grounded LVQA benchmarks (CG-Bench, Haystack-LVBench, Haystack-Ego4D), VTS outperforms the strongest prior agentic methods by +12.5 mIoU on CG-Bench and +7.4 T-F1 on Haystack-Ego4D. The learned policy also transfers to general long-video QA, surpassing all prior agentic baselines on Video-MME, MLVU, and LVBench by up to +7.1 accuracy points. Ablations confirm that self-correcting hierarchical search is the central mechanism behind these gains: removing either adaptive descent or explicit backtracking substantially degrades performance. Code is available at https://github.com/CeeZh/VTS.
Ce Zhang, Ziyang Wang, Yulu Pan et al.· 0 citations
Known for his clear and elegant writing style, Bertsekas shaped fields from control and optimization to large-scale computation and artificial intelligence.
