AI Paper Atlas

Research map / Video / Video-language understanding

Learning to Watch: Active Video Anomaly Understanding via Interleaved Policy Optimization

Mengjingcheng Mo, Jiaxu Leng, Xinbo Gao

ICML 2026 · Video › Video-language understanding · approach: Event and anomaly reasoning in video · 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 active closed-loop interaction instead of uniform-sampling one-shot pipeline (ablation)Anom-π · XD-ViolenceAP (%): 65.99 -> 72.29Improved
add active interaction instead of random-sampling one-shot pipeline (ablation)Anom-π · XD-ViolenceAP (%): 52.54 -> 72.29Improved
train with iDPO rather than vanilla DPO without curiosity (AEI) (ablation)Anom-π · XD-ViolenceAP (%): 66.57 -> 72.29Improved
train with iDPO rather than positive-only SFT (ablation)Anom-π · XD-ViolenceAP (%): 56.86 -> 72.29Improved
add iDPO policy training over zero-shot model (ablation)Anom-π · XD-ViolenceAP (%): 34.92 -> 72.29Improved
add Think action (ablation)Anom-π · XD-ViolenceAP (%): 65.54 -> 72.29Improved
add Sample action (ablation)Anom-π · XD-ViolenceAP (%): 67.83 -> 72.29Improved
add Backtrack action (ablation)Anom-π · XD-ViolenceAP (%): 70.49 -> 72.29Improved
add Expand action (ablation)Anom-π · XD-ViolenceAP (%): 71.48 -> 72.29Improved
score trajectories with interaction cost, not final prediction only (ablation)iDPO · XD-ViolenceAP (%): -5.42 AP when scoring by final prediction only; -2.25 when removing interaction costImproved
set interaction-cost weight eta to 0.5 instead of 0 (no cost) (ablation)iDPO · XD-ViolenceAP (%): 70.04 -> 72.29Improved
increase eta to 2.0 from 0.5 (ablation)iDPO · XD-ViolenceAP (%): 72.29 -> 69.90Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 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