AI Paper Atlas

Research map / Video / Video segmentation and tracking

Cross-Domain Learning for Video Anomaly Detection with Limited Supervision

Yashika Jain, Ali Dabouei, Min Xu

ECCV 2024 · Video › Video segmentation and tracking · approach: Weakly supervised video anomaly detection · 7 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 unlabeled external data without uncertainty weighting or cosine loss (ablation)Weakly-supervised VAD baseline · UCF-R open-set c=1AUC (%): 84.32 -> 84.67Improved
add uncertainty-based reweighing of external-data bias (ablation)CDL with external data · UCF-R open-set c=1AUC (%): 84.67 -> 84.80Improved
add cosine similarity loss term on I3D/CLIP representations (ablation)CDL with uncertainty weighting · UCF-R open-set c=1AUC (%): 84.80 -> 85.39Improved
add CDL with HACS external dataOurs trained on UCF-Crime · XDV (source: UCF-Crime)AP (%): 58.13 -> 65.14Improved
add CDL with XDV as external dataOurs trained on UCF-Crime · XDV (source: UCF-Crime)AP (%): 58.13 -> 68.37Improved
add external data to source-domain trainingOurs trained on UCF-Crime · UCF-Crime original testAUC (%): 84.49 -> 84.73No clear effect
add CDL with HACS external dataOurs trained on XDV · UCF-R (source: XDV)AUC (%): 76.39 -> 88.50Improved
add CDL external data to model trained on XDVOurs trained on XDV · XDV (source: XDV)AP (%): 75.13 -> 78.61Improved
add CDL unlabeled data in open-set setting across c values (ablation)Ours · UCF-Crime open-setAUC (%): 75.17 -> 77.45 (c=1); 83.02 -> 83.37 (c=9)Improved
increase cosine similarity loss coefficient lambda_3 from 1e-9 to 1e-3, then beyond (ablation)CDL · UCF-Crime open-set c=1AUC (%)Mixed
increase number of segments n_s from 16 to 64, then beyond (ablation)CDL · UCF-Crime open-set c=1AUC (%)Mixed
add more unlabeled HACS videos beyond 11,000 (ablation)CDL (UCF-Crime source) · XDVAP (%)No clear effect

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