{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,5,4]],"date-time":"2026-05-04T10:14:40Z","timestamp":1777889680611,"version":"3.51.4"},"reference-count":62,"publisher":"IEEE","license":[{"start":{"date-parts":[[2025,10,19]],"date-time":"2025-10-19T00:00:00Z","timestamp":1760832000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-029"},{"start":{"date-parts":[[2025,10,19]],"date-time":"2025-10-19T00:00:00Z","timestamp":1760832000000},"content-version":"stm-asf","delay-in-days":0,"URL":"https:\/\/doi.org\/10.15223\/policy-037"}],"funder":[{"DOI":"10.13039\/501100003725","name":"National Research Foundation of Korea","doi-asserted-by":"publisher","id":[{"id":"10.13039\/501100003725","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100014188","name":"MSIT","doi-asserted-by":"publisher","award":["RS-2025-00554790"],"award-info":[{"award-number":["RS-2025-00554790"]}],"id":[{"id":"10.13039\/501100014188","id-type":"DOI","asserted-by":"publisher"}]},{"DOI":"10.13039\/501100002573","name":"Yonsei University","doi-asserted-by":"publisher","award":["RS-2020-11201361"],"award-info":[{"award-number":["RS-2020-11201361"]}],"id":[{"id":"10.13039\/501100002573","id-type":"DOI","asserted-by":"publisher"}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":[],"published-print":{"date-parts":[[2025,10,19]]},"DOI":"10.1109\/iccv51701.2025.02224","type":"proceedings-article","created":{"date-parts":[[2026,4,29]],"date-time":"2026-04-29T19:45:49Z","timestamp":1777491949000},"page":"23990-24000","source":"Crossref","is-referenced-by-count":0,"title":["Multi-Granular Spatio-Temporal Token Merging for Training-Free Acceleration of Video LLMs"],"prefix":"10.1109","author":[{"given":"Jeongseok","family":"Hyun","sequence":"first","affiliation":[{"name":"Yonsei University"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Sukjun","family":"Hwang","sequence":"additional","affiliation":[{"name":"Carnegie Mellon University"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Su Ho","family":"Han","sequence":"additional","affiliation":[{"name":"Yonsei University"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Taeoh","family":"Kim","sequence":"additional","affiliation":[{"name":"NAVER Cloud"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Inwoong","family":"Lee","sequence":"additional","affiliation":[{"name":"NAVER Cloud"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Dongyoon","family":"Wee","sequence":"additional","affiliation":[{"name":"NAVER Cloud"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Joon-Young","family":"Lee","sequence":"additional","affiliation":[{"name":"Adobe Research"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Seon Joo","family":"Kim","sequence":"additional","affiliation":[{"name":"Yonsei University"}],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Minho","family":"Shim","sequence":"additional","affiliation":[{"name":"NAVER Cloud"}],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"263","reference":[{"key":"ref1","article-title":"Qwen-vl: A frontier large vision-language model with versatile abilities","author":"Bai","year":"2023","journal-title":"arXiv preprint"},{"key":"ref2","article-title":"Token merging: Your vit but faster","author":"Bolya","year":"2023","journal-title":"ICLR"},{"key":"ref3","article-title":"Pyramidkv: Dynamic kv cache compression based on pyramidal information funneling","author":"Cai","year":"2024","journal-title":"arXiv preprint"},{"key":"ref4","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2017.502"},{"key":"ref5","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52733.2024.01311"},{"key":"ref6","doi-asserted-by":"publisher","DOI":"10.1007\/978-3-031-73004-7_2"},{"key":"ref7","article-title":"Longvila: Scaling long-context visual language models for long videos","author":"Chen","year":"2025","journal-title":"ICLR"},{"key":"ref8","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52733.2024.01776"},{"key":"ref9","article-title":"FlashAttention-2: Faster attention with better parallelism and work partitioning","author":"Dao","year":"2024","journal-title":"ICLR"},{"key":"ref10","doi-asserted-by":"publisher","DOI":"10.52202\/068431-1189"},{"key":"ref11","article-title":"The llama 3 herd of models","author":"Dubey","year":"2024","journal-title":"arXiv preprint"},{"key":"ref12","first-page":"35946","article-title":"Masked autoencoders as spatiotemporal learners","volume":"35","author":"Feichtenhofer","year":"2022","journal-title":"NeurIPS"},{"key":"ref13","doi-asserted-by":"publisher","DOI":"10.1007\/BF00288933"},{"key":"ref14","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52734.2025.02245"},{"key":"ref15","article-title":"Framefusion: Combining similarity and importance for video token reduction on large visual language models","author":"Fu","year":"2024","journal-title":"arXiv preprint"},{"key":"ref16","volume-title":"Google. Caching - google ai","year":"2024"},{"key":"ref17","article-title":"Gemini Team Google. Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context","year":"2024","journal-title":"arXiv preprint"},{"key":"ref18","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR.2016.90"},{"key":"ref19","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52688.2022.01553"},{"key":"ref20","article-title":"Prunevid: Visual token pruning for efficient video large language models","author":"Huang","year":"2024","journal-title":"arXiv preprint"},{"key":"ref21","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52733.2024.01300"},{"key":"ref22","volume-title":"Needle in a haystack - pressure testing llms","author":"Kamradt","year":"2023"},{"key":"ref23","article-title":"Handwritten digit recognition with a backpropagation network","author":"LeCun","year":"1989","journal-title":"NeurIPS"},{"key":"ref24","first-page":"12888","article-title":"Blip: Bootstrapping language-image pre-training for unified vision-language understanding and generation","author":"Li","year":"2022","journal-title":"ICML"},{"key":"ref25","first-page":"19730","article-title":"Blip-2: Bootstrapping language-image pre-training with frozen image encoders and large language models","author":"Li","year":"2023","journal-title":"ICML"},{"key":"ref26","doi-asserted-by":"publisher","DOI":"10.1007\/s11432-024-4321-9"},{"key":"ref27","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52733.2024.02095"},{"key":"ref28","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.emnlp-main.342"},{"key":"ref29","doi-asserted-by":"publisher","DOI":"10.52202\/075280-1516"},{"key":"ref30","article-title":"World model on million-length video and language with blockwise ringattention","author":"Liu","year":"2024","journal-title":"arXiv preprint"},{"key":"ref31","article-title":"Ring attention with blockwise transformers for near-infinite context","author":"Liu","year":"2024","journal-title":"ICLR"},{"key":"ref32","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2024.acl-long.679"},{"key":"ref33","first-page":"46212","article-title":"Egoschema: A diagnostic benchmark for very longform video language understanding","volume":"36","author":"Mangalam","year":"2023","journal-title":"NeurIPS"},{"key":"ref34","first-page":"606","article-title":"Efficiently scaling transformer inference","volume":"5","author":"Pope","year":"2023","journal-title":"MLSys"},{"key":"ref35","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52733.2024.01357"},{"key":"ref36","article-title":"xgen-mm-vid (blip-3video): You only need 32 tokens to represent a video even in vlms","author":"Ryoo","year":"2024","journal-title":"arXiv preprint"},{"key":"ref37","doi-asserted-by":"publisher","DOI":"10.1145\/356924.356930"},{"key":"ref38","article-title":"Llava-prumerge: Adaptive token reduction for efficient large multimodal models","author":"Shang","year":"2024","journal-title":"arXiv preprint"},{"key":"ref39","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52733.2024.01725"},{"key":"ref40","doi-asserted-by":"publisher","DOI":"10.1016\/j.neucom.2023.127063"},{"key":"ref41","doi-asserted-by":"publisher","DOI":"10.1109\/83.287030"},{"key":"ref42","doi-asserted-by":"publisher","DOI":"10.1109\/TCSVT.2012.2221191"},{"key":"ref43","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52734.2025.01769"},{"key":"ref44","doi-asserted-by":"publisher","DOI":"10.1145\/321879.321884"},{"key":"ref45","article-title":"OpenAI Team. Gpt-4o system card","year":"2024","journal-title":"arXiv preprint"},{"key":"ref46","first-page":"10078","article-title":"Videomae: Masked autoencoders are data-efficient learners for self-supervised video pre-training","volume":"35","author":"Tong","year":"2022","journal-title":"NeurIPS"},{"key":"ref47","article-title":"Llama: Open and efficient foundation language models","author":"Touvron","year":"2023","journal-title":"arXiv preprint"},{"key":"ref48","doi-asserted-by":"publisher","DOI":"10.1109\/ICCV.2015.510"},{"key":"ref49","article-title":"Look-m: Lookonce optimization in kv cache for efficient multimodal longcontext inference","author":"Wan","year":"2024","journal-title":"arXiv preprint"},{"key":"ref50","article-title":"Qwen2-vl: Enhancing vision-language model\u2019s perception of the world at any resolution","author":"Wang","year":"2024","journal-title":"arXiv preprint"},{"key":"ref51","article-title":"Videollamb: Long-context video understanding with recurrent memory bridges","author":"Wang","year":"2024","journal-title":"arXiv preprint"},{"key":"ref52","doi-asserted-by":"publisher","DOI":"10.1109\/TCSVT.2003.815165"},{"key":"ref53","first-page":"28828","article-title":"Longvideobench: A benchmark for long-context interleaved video-language understanding","volume":"37","author":"Wu","year":"2025","journal-title":"NeurIPS"},{"key":"ref54","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR46437.2021.00965"},{"key":"ref55","article-title":"Pyramiddrop: Accelerating your large vision-language models via pyramid visual redundancy reduction","author":"Xing","year":"2025","journal-title":"CVPR"},{"key":"ref56","article-title":"Qwen2 technical report","author":"Yang","year":"2024","journal-title":"arXiv preprint"},{"key":"ref57","article-title":"Pyramidinfer: Pyramid kv cache compression for high-throughput 11 m inference","author":"Yang","year":"2024","journal-title":"arXiv preprint"},{"key":"ref58","doi-asserted-by":"publisher","DOI":"10.18653\/v1\/2023.emnlp-demo.49"},{"key":"ref59","article-title":"Long context transfer from language to vision","author":"Zhang","year":"2024","journal-title":"arXiv preprint"},{"key":"ref60","article-title":"Video instruction tuning with synthetic data","author":"Zhang","year":"2024","journal-title":"arXiv preprint"},{"key":"ref61","article-title":"Needle in a video haystack: A scalable synthetic evaluator for video mllms","author":"Zhao","year":"2024","journal-title":"arXiv preprint"},{"key":"ref62","doi-asserted-by":"publisher","DOI":"10.1109\/CVPR52734.2025.01278"}],"event":{"name":"2025 IEEE\/CVF International Conference on Computer Vision (ICCV)","location":"Honolulu, HI, USA","start":{"date-parts":[[2025,10,19]]},"end":{"date-parts":[[2025,10,25]]}},"container-title":["2025 IEEE\/CVF International Conference on Computer Vision (ICCV)"],"original-title":[],"link":[{"URL":"http:\/\/xplorestaging.ieee.org\/ielx8\/11443115\/11443287\/11446256.pdf?arnumber=11446256","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,5,1]],"date-time":"2026-05-01T05:14:36Z","timestamp":1777612476000},"score":1,"resource":{"primary":{"URL":"https:\/\/ieeexplore.ieee.org\/document\/11446256\/"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2025,10,19]]},"references-count":62,"URL":"https:\/\/doi.org\/10.1109\/iccv51701.2025.02224","relation":{},"subject":[],"published":{"date-parts":[[2025,10,19]]}}}