AI Paper Atlas

Research map / Video / Video segmentation and tracking

End-to-End Compressed Video Representation Learning for Generic Event Boundary Detection

Congcong Li, Xinyao Wang, Longyin Wen, Dexiang Hong, Tiejian Luo, Libo Zhang

CVPR 2022 · Video › Video segmentation and tracking · approach: Self-supervised video representation learning · 18 citations (OpenAlex)

Check 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 the end-to-end frame feature strategy rather than repeated PC context-window inputs (ablation)PC · Kinetics-GEBD local minvalF1: 0.621 -> 0.634Improved
Add Gaussian-smoothed soft labels (ablation)PC + E2E · Kinetics-GEBD local minvalF1: 0.634 -> 0.654Improved
Add optical flow (ablation)PC · Kinetics-GEBD local minvalF1: 0.621 -> 0.646Improved
Use SCCE to fuse I-frame features with compressed P-frame information (ablation)PC · Kinetics-GEBD local minvalF1: 0.621 -> 0.669Improved
Add optical flow (ablation)Ours · Kinetics-GEBD local minvalF1: 0.654 -> 0.661Improved
Use SCCE rather than a vanilla compressed-feature encoder (ablation)Ours · Kinetics-GEBD local minvalF1: 0.657 -> 0.686Improved
Add the temporal contrastive module with window size k=2 (ablation)Ours · Kinetics-GEBD local minvalF1: 0.686 -> 0.710Improved
Set temporal contrastive window size to k=8 instead of k=2 (ablation)Ours · Kinetics-GEBD local minvalF1: 0.710 -> 0.746Improved
Increase the temporal contrastive window beyond k=8 (ablation)Ours · Kinetics-GEBD local minvalrecall: 0.757 (k=8) -> 0.725 (k=10); 0.696 (k=12)Mixed
Use the proposed compressed-domain representationOurs versus PC + Optical Flow · Kinetics-GEBD val, avg Rel.Dis.F1: 0.830 -> 0.865Improved
Use SCCE-based compressed features instead of vanilla compressed featuresOurs · UCF-101accuracyImproved

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