{"total":15,"items":[{"citing_arxiv_id":"2606.25658","ref_index":28,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Towards a Dynamic and Fixed-budget Memory Bank for Efficient Streaming Video Understanding","primary_cat":"cs.CV","submitted_at":"2026-06-24T10:11:08+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"CausalMem constructs a dynamic fixed-budget memory bank for streaming video in MLLMs via online semantic basis updates, achieving 20x token compression and accuracy gains on benchmarks when applied to LLaVA-OneVision and Qwen2.5-VL.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.09547","ref_index":26,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Streaming Interventions: Can Video Large Language Models Correct Mistakes as They Occur?","primary_cat":"cs.CV","submitted_at":"2026-06-08T14:27:20+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"Introduces Ego-MC-Bench benchmark and Ego-CoMist synthetic dataset showing that fine-tuning video LLMs on proactive mistake corrections improves performance especially for smaller models.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2606.07433","ref_index":174,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Watch, Remember, Reason: Human-View Video Understanding with MLLMs","primary_cat":"cs.CV","submitted_at":"2026-06-05T16:29:13+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":4.0,"formal_verification":"none","one_line_summary":"This is a survey that frames video MLLM research via a human-view formulation of perceptual representations, memory states, reasoning traces, and predictions, then reviews methods, datasets, benchmarks, and open problems.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"decide, and react to dynamic visual stimuli. IEEE TRANSACTIONS ON PATTERN ANAL YSIS AND MACHINE INTELLIGENCE 12 TABLE 4: Representative works aboutHow to Reason?(Sec. 3.3). Method Year/Conf. Training Highlight Section 3.3.1: Text-only Reasoning (Agentic) VideoAgent [141] ECCV 2024 Training-free Agentic use of temporal/object memory with iterative tool-based reasoning DoraemonGPT [174] ICML 2024 Training-free Dynamic spatio-temporal memory with Monte-Carlo Tree Search for multi-step explanations Video-of-Thought [175] ICML 2024 SFT Video-of-Thought framework for multi-stage perception-to-cognition reasoning VCA [176] ICCV 2025 Training-free Curiosity-driven agent that adaptively explores frames via tree search Flow4Agent [177] ICCV 2025 Training-free Optical-flow motion priors for adaptive temporal granularity and evidence focusing"},{"citing_arxiv_id":"2605.31598","ref_index":42,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Linear Scaling Video VLMs for Long Video Understanding","primary_cat":"cs.CV","submitted_at":"2026-05-29T17:59:02+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"StateKV is an inference-time technique that replaces quadratic self-attention prefill in video VLMs with a fixed-capacity importance-based recurrent state, keeping accuracy near full attention on long-video benchmarks without retraining.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.25621","ref_index":23,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"StreamOV: Streaming Omni-Video Understanding via Evidence-Guided Memory and Response Triggering","primary_cat":"cs.CV","submitted_at":"2026-05-25T09:23:19+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"StreamOV proposes evidence-guided long-short term memory and a hidden-state-driven trigger for efficient online audio-visual reasoning in streaming videos, along with the SOVBench benchmark for multi-turn evaluation.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.22269","ref_index":29,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"MuKV: Multi-Grained KV Cache Compression for Long Streaming Video Question-Answering","primary_cat":"cs.CV","submitted_at":"2026-05-21T10:13:03+00:00","verdict":"CONDITIONAL","verdict_confidence":"MODERATE","novelty_score":5.0,"formal_verification":"none","one_line_summary":"MuKV adds multi-grained KV cache compression at patch-frame-segment levels plus semi-hierarchical retrieval to raise accuracy and cut memory in long video question-answering.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.17921","ref_index":12,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"An Efficient Streaming Video Understanding Framework with Agentic Control","primary_cat":"cs.CV","submitted_at":"2026-05-18T06:29:44+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"R3-Streaming uses cascaded control with age-aware memory forgetting and TB-GRPO reinforcement learning to reach SOTA scores of 57.92 on OVO-Bench and 76.36 on StreamingBench with 95-96% fewer visual tokens.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2605.07897","ref_index":21,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Semantic-Aware Adaptive Visual Memory for Streaming Video Understanding","primary_cat":"cs.CV","submitted_at":"2026-05-08T15:40:40+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"SAVEMem improves streaming video understanding scores by adding semantic awareness to memory compression and query-adaptive retrieval without any model training.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"To meet these requirements under a finite budget, existing work mainly focuses on compressing visual tokens before they enter the LLM [2, 37, 40, 41] via visual similarity heuristics, with FluxMem [37] as a strong recent example using a hierarchical memory governed by adaptive Otsu-based thresholds [23]. Another line of work manages the KV cache duringtheprefillstagetocopewithunpredictablequeries: ReKV[ 6]andLiveVLM[ 21]retrievequery-relevantKV-cache entries at inference, while WeaveTime [47] triggers coarse-to-fine recall via prediction uncertainty. SimpleStream [27] further shows that feeding only the last𝑁 frames to an off-the-shelf VLM is already competitive on many benchmarks, hinting at a perception-memory trade-off between present- and past-oriented queries."},{"citing_arxiv_id":"2605.01858","ref_index":10,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Decouple and Cache: KV Cache Construction for Streaming Video Understanding","primary_cat":"cs.CV","submitted_at":"2026-05-03T13:02:44+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":5.0,"formal_verification":"none","one_line_summary":"DSCache decouples cumulative past and instant KV caches with position-agnostic encoding to adapt offline VideoVLLMs to streaming video, delivering 2.5% average accuracy gains on QA benchmarks.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.24317","ref_index":41,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Don't Pause! Every prediction matters in a streaming video","primary_cat":"cs.CV","submitted_at":"2026-04-27T11:07:03+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"SPOT-Bench tests real-time streaming video perception with timeliness metrics, exposing limitations in current models and introducing AsynKV as an improved baseline.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"chatgpt: Towards detailed video understanding via large vision and language models. arXiv preprint arXiv:2306.05424, 2023. 1, 4 21 SPOT-Bench Technical Report [40] Karttikeya Mangalam, Raiymbek Akshulakov, and Jitendra Malik. Egoschema: A diagnostic benchmark for very long-form video language understanding.Advances in Neural Information Processing Systems, 36:46212-46244, 2023. 1 [41] Zhenyu Ning, Guangda Liu, Qihao Jin, Wenchao Ding, Minyi Guo, and Jieru Zhao. Livevlm: Efficient online video understanding via streaming-oriented kv cache and retrieval.arXiv preprint arXiv:2505.15269, 2025. 2, 9 [42] Junbo Niu, Yifei Li, Ziyang Miao, Chunjiang Ge, Yuanhang Zhou, Qihao He, Xiaoyi Dong, Haodong Duan, Shuangrui Ding, Rui Qian, et al."},{"citing_arxiv_id":"2604.11411","ref_index":32,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Online Reasoning Video Object Segmentation","primary_cat":"cs.CV","submitted_at":"2026-04-13T12:55:56+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"The work introduces the ORVOS task, the ORVOSB benchmark with causal annotations across 210 videos, and a baseline using updated prompts plus a temporal token reservoir.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2604.10060","ref_index":16,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"Mosaic: Cross-Modal Clustering for Efficient Video Understanding","primary_cat":"cs.PF","submitted_at":"2026-04-11T06:54:56+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"Mosaic uses cross-modal clusters as the unit for KVCache organization in VLMs to achieve up to 1.38x speedup in streaming long-video understanding.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"When these steps are performed strictly in sequence, data transfer repeatedly falls on the critical path, causing GPU com- putation to stall on I/O and limiting overall system throughput. MOSAICis motivated by an important observation in trans- former decoding: query embeddings from adjacent layers are often highly similar. This phenomenon is largely induced by residual connections [16], which preserve a substantial portion of the representation across layers. As a result, neighboring layers tend to retrieve highly overlapping sets of clusters. This cross-layer similarity makes it possible to use the query of the current layer as a reliable proxy for predicting which clusters will be needed in the next layer. MOSAICleverages this property to overlap computation"},{"citing_arxiv_id":"2604.06036","ref_index":47,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"CodecSight: Leveraging Video Codec Signals for Efficient Streaming VLM Inference","primary_cat":"cs.DC","submitted_at":"2026-04-07T16:31:45+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"CodecSight reuses video codec signals for online patch pruning before the vision transformer and selective KV-cache refresh in the LLM, delivering up to 3x higher throughput and 87% lower GPU compute than prior baselines with 0-8% F1 drop.","context_count":1,"top_context_role":"background","top_context_polarity":"background","context_text":"Accessed 2026-03-26. [45] NVIDIA. 2026. VDEC Application Note. https://docs.nvidia. com/video-technologies/video-codec-sdk/13.0/nvdec-application- note/index.html [46] Tsung-Yin Ou, Andrés Ponce, Cody Lee, and Areoll Wu. 2025. Real- time retail planogram compliance application using computer vision and virtual shelves.Scientific Reports15, 1 (2025), 43898. [47] Junting Pan, Ziyi Lin, Xiatian Zhu, Jing Shao, and Hongsheng Li. 2022. ST-Adapter: Parameter-Efficient Image-to-Video Transfer Learning.. InThe 36th Annual Conference on Neural Information Processing Sys- tems (NeurIPS). 14 CodecSight: Leveraging Video Codec Signals for Efficient Streaming VLM Inference Conference'17, July 2017, Washington, DC, USA"},{"citing_arxiv_id":"2603.20284","ref_index":20,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"STAC: Plug-and-Play Spatio-Temporal Aware Cache Compression for Streaming 3D Reconstruction","primary_cat":"cs.CV","submitted_at":"2026-03-18T06:36:46+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":7.0,"formal_verification":"none","one_line_summary":"STAC compresses KV caches in streaming 3D reconstruction transformers via temporal token preservation with decayed attention, spatial voxel compression, and chunked multi-frame optimization, delivering 10x memory reduction and 4x faster inference at SOTA quality.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null},{"citing_arxiv_id":"2601.14724","ref_index":32,"ref_count":1,"confidence":0.98,"is_internal_anchor":true,"paper_title":"HERMES: KV Cache as Hierarchical Memory for Efficient Streaming Video Understanding","primary_cat":"cs.CV","submitted_at":"2026-01-21T07:26:15+00:00","verdict":"UNVERDICTED","verdict_confidence":"LOW","novelty_score":6.0,"formal_verification":"none","one_line_summary":"HERMES organizes the KV cache into a hierarchical memory to enable real-time streaming video understanding in MLLMs, achieving 10x faster TTFT and up to 11.4% accuracy gains on streaming benchmarks with 68% fewer tokens.","context_count":0,"top_context_role":null,"top_context_polarity":null,"context_text":null}],"limit":50,"offset":0}