AI Paper Atlas

Research map / Video / Action recognition

Seeing Motion Through Polarity for Event-based Action Recognition

Meiqi Cao, Jiachao Zhang, Xin Jiang, Rui Yan, Yazhou Yao, Zechao Li, Xiangbo Shu

CVPR 2026 · Video › Action recognition · approach: Disentangled action recognition

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 POKER (PMC + PMR) to EventBind with stacked framesEventBind · SeActtop-1 accuracy (%): 67.24 -> 69.82Improved
add POKER to EventBind with reconstructed framesEventBind · THUE-ACT-50-CHLtop-1 accuracy (%): 61.14 -> 63.72Improved
add POKER to EventBind with reconstructed frames on DVS ActionEventBind · DVS Actiontop-1 accuracy (%): 98.24 -> 99.60Improved
add Polarity Motion Capturer (PMC) (ablation)EventBind · SeActtop-1 accuracy (%): 74.13 -> 75.00Improved
add Polarity Motion Capturer (PMC) (ablation)EventBind · THUE-ACT-50-CHLtop-1 accuracy (%): 61.14 -> 62.24Improved
add Polarity Motion Reasoner (PMR) on top of PMC (ablation)EventBind + PMC · SeActtop-1 accuracy (%): 75.00 -> 76.72Improved
add PMR on top of PMC (ablation)EventBind + PMC · THUE-ACT-50-CHLtop-1 accuracy (%): 62.24 -> 63.72Improved
use learnable router fusion instead of concatenation in PMC (ablation)PMC · THUE-ACT-50-CHLaccuracy (%): 61.51 -> 63.72Improved
use learnable router fusion instead of element-wise addition (ablation)PMC · THUE-ACT-50-CHLaccuracy (%): 63.35 -> 63.72Improved
use polarity alignment loss instead of standard contrastive loss (ablation)POKER · THUE-ACT-50-CHLaccuracy (%): 62.62 -> 63.72Improved
use Gemini-2.5-pro rather than Qwen3-VL-30B as MLLM in PMR (ablation)PMR · SeActtop-1 accuracy (%): 75.00 -> 76.72Improved
use general-purpose MLLMs zero-shot instead of polarity-guided pipeline (ablation)MLLMs · DVS Action zero-shottop-1 accuracy (%): 99.6 vs 51.7 (GPT-4o-mini)Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 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