AI Paper Atlas

Research map / Video / Video segmentation and tracking

Beyond Pixels: Mining Compressed Domain Artifacts for Efficient AI-Generated Video Detection

Anran Zhu, Zhengli Shi, Chende Zheng, Chenhao Lin, Zhengyu Zhao, Le Yang, Chong Zhang, Shuai Liu, Chao Shen

ICML 2026 · Video › Video segmentation and tracking · approach: Self-supervised video representation learning

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
fuse I-frame, motion vector and residual branches instead of I-frame only (ablation)STREAM · GVFmAP: 0.885 -> 0.950Improved
use compressed-domain bitstream inputs instead of pixel-domain decodingAIGV detector · inference time per 1,000 videosseconds: 40 -> 16 (vs D3)Improved
use compressed-domain STREAM instead of best pixel-domain baseline D3AIGV detection · 5-dataset averagemAP: 0.943 -> 0.965Improved
use ResNet-152 instead of ResNet-18 backbone (ablation)STREAM · 10 generatorsmAP: 0.9149 -> 0.9204No clear effect
add strong Gaussian blur to test videos (residual only) (ablation)STREAM residual branch · 10 generatorsmAP: 0.941 -> 0.842Hurt
add strong Gaussian blur to test videos (MV only) (ablation)STREAM MV branch · 10 generatorsmAP: 0.937 -> 0.924Hurt
fuse all three modalities under strong Gaussian blur (ablation)STREAM · 10 generatorsmAP: 0.975 -> 0.949Hurt
increase rounds of loop-filter re-encoding from 0 to 6STREAM, AIGVDet, NPR · multi-round re-encodingmAP: 0.976 -> 0.697 (STREAM)Hurt
H.264 re-encoding at QP=35 of test videosSTREAM · VideoPhy/GenVideo/GVFavg mAP: 0.959 -> 0.902Hurt
overlay 50% opacity watermark on test videosSTREAM · VideoPhy/GenVideo/GVFavg mAP: 0.959 -> 0.827Hurt
change resolution to 720p or re-encode with H.265STREAM · VideoPhy/GenVideo/GVFavg mAP: 0.959 -> 0.958No 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