AI Paper Atlas

Research map / Video / Video-language understanding

Keep It in Mind: User Centric Continual Spatial Intelligence Reasoning in Egocentric Video Streams

Yun wang, Junbin Xiao, Han Lyu, Yifan Wang, Jing Zuo, Zhanjie Zhang, Hong Huang, Dapeng Wu, Angela Yao

ICML 2026 · Video › Video-language understanding · approach: Egocentric video understanding · 0 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
add per-frame MLLM spatial descriptions as extra context (ablation)Qwen3-VL-8B · UCS-Benchoverall accuracy: 44.0 -> 44.5No clear effect
add pose-anchored scene graph memory (ablation)Qwen3-VL-8B · UCS-Benchoverall accuracy: 44.0 -> 50.2Improved
add spatial descriptions on top of scene graph (ablation)DirectMe · UCS-Benchoverall accuracy: 50.2 -> 50.5No clear effect
wrap InternVL3-8B with DirectMeInternVL3-8B · UCS-Benchoverall accuracy: 43.8 -> 49.7Improved
wrap Qwen3-VL-8B with DirectMeQwen3-VL-8B · UCS-Benchoverall accuracy: 44.0 -> 50.5Improved
use spatial-centric models instead of general MLLMsQwen3-VL-8B baseline · UCS-Benchoverall accuracy: SenseNova-SI 37.8, Spatial-MLLM 34.3 vs Qwen3-VL-8B 44.0Hurt
use streaming-specific architectures instead of general MLLMsvideo QA baselines · UCS-Benchoverall accuracy: Flash-VStream 28.5, Dispider 36.7, VideoChat-Online 37.5 vs InternVL3-8B 43.8Hurt
scale InternVL3 from 8B to 38BInternVL3 · UCS-Benchoverall accuracy: 43.8 -> 49.2Improved
scale Qwen3-VL from 8B to 32BQwen3-VL · UCS-Benchoverall accuracy: 44.0 -> 49.0Improved
scale InternVL3.5 from 8B to 38BInternVL3.5 · UCS-Benchoverall accuracy: 42.9 -> 48.1Improved
increase evidence-to-question time gapall models including DirectMe · UCS-BenchaccuracyHurt

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