AI Paper Atlas

Research map / Video / Video generation

I'm a Map! Interpretable Motion-Attentive Maps: Spatio-Temporally Localizing Concepts in Video Diffusion Transformers

Youngjun Jun, Seil Kang, Woojung Han, Seong Jae Hwang

CVPR 2026 · Video › Video generation · approach: Efficient video diffusion models

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
use IMAP instead of ConceptAttentionCogVideoX-5B · MeViS motion localizationMLS average: 0.45 -> 0.62Improved
use GramCol instead of ConceptAttentionCogVideoX-5B · VSPW zero-shot video semantic segmentationmIoU: 25.0 -> 28.9Improved
use GramCol instead of ConceptAttentionCogVideoX-5B · VSPW zero-shot video semantic segmentationmVC8: 80.4 -> 75.2Hurt
add AnyUp upsampling to saliency mapsGramCol · VSPW zero-shot video semantic segmentationmIoU: 28.9 -> 30.1Improved
replace ConceptAttention with GramCol (no layer or head selection) (ablation)CogVideoX-5B · MeViS motion localizationMLS average: 0.37 -> 0.41Improved
add lambda2-based layer selection (ablation)GramCol · MeViS motion localizationMLS average: 0.41 -> 0.46Improved
add motion head selection (CHI top-5) (ablation)GramCol · MeViS motion localizationMLS average: 0.41 -> 0.47Improved
combine layer selection and motion head selection (ablation)GramCol · MeViS motion localizationMLS average: 0.41 -> 0.62Improved
apply softmax over concepts to IMAP (ablation)IMAP · MeViS motion localizationMLS average: 0.62 -> 0.60Mixed
select heads with higher Calinski-Harabasz separation score (ablation)GramCol attention heads · MeViS subset, CogVideoX-2BMLS: Pearson r = 0.60Improved

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