AI Paper Atlas

Research map / Video / Video-language understanding

Thinking in Streaming Video

Zikang Liu, Longteng Guo, Handong Li, Ru Zhen, Xingjian He, Ruyi Ji, Xiaoming Ren, Yanhao Zhang, Haonan Lu, Jing Liu

ECCV 2026 · Video › Video-language understanding · approach: Long video understanding · 22 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
increase reasoning token budget from 0 to 20 tokens per second (ablation)ThinkStream-3B · OVO-Bench Backwardaccuracy: 41.8 -> 52.3Improved
raise reasoning budget from 20 to 30 tokens per second (ablation)ThinkStream-3B · OVO-Bench Backwardaccuracy: 52.3 -> 52.6, latency 380 ms -> 505 msNo clear effect
use 20s instead of 5s visual KV window (ablation)ThinkStream-3B · OVO-Bench Backwardaccuracy: 46.7 -> 52.3Improved
expand visual window from 20s to 30s (ablation)ThinkStream-3B · OVO-Bench Backwardaccuracy: 52.3 -> 51.6Hurt
add cold-start CoT memory versus no memory (ablation)Qwen2.5-VL-3B streaming · Memory variants averageaverage score: 56.9 -> 60.5Improved
use discrete caption tokens as memory instead of no memory (ablation)Qwen2.5-VL-3B streaming · Memory variants averageaverage score: 56.9 -> 48.7Hurt
add RLVR post-training on top of cold-start CoT memory (ablation)ThinkStream-3B · Memory variants averageaverage score: 60.5 -> 64.8Improved
apply ThinkStream framework to Qwen2.5-VL-3BQwen2.5-VL-3B · OVO-Benchoverall average accuracy: 51.00 -> 59.66Improved
apply ThinkStream framework to Qwen2.5-VL-7BQwen2.5-VL-7B · OVO-Bench Backward Tracingaverage accuracy: 52.28 -> 60.68Improved
apply ThinkStream with visual token eviction to offline benchmarksQwen2.5-VL-3B · VideoMME and LongVideoBenchaccuracy: 61.5 -> 61.9 (VideoMME), 54.2 -> 56.4 (LongVideoBench)Improved
CUDA Graph streaming inference engine versus standard Qwen2.5-VL-3B decoding at batch size 1ThinkStream-3B · Token decoding speedtokens/s: 30.06 -> 154.07Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy