AI Paper Atlas

Research map / Detection and segmentation / CNN features

Image Manipulation Detection by Multi-View Multi-Scale Supervision

Xinru Chen, Chengbo Dong, Jiaqi Ji, Juan Cao, Xirong Li

ICCV 2021 · Detection and segmentation › CNN features · approach: Forgery detection · 304 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 image-scale classification loss to Seg (ablation)Seg · DEFACTO-12kpixel-level mean F1: 0.546 -> 0.463Hurt
add image-scale classification loss to Seg (ablation)Seg · DEFACTO-12kimage-level specificity: 0.620 -> 0.778Improved
add image-scale classification loss to Seg (ablation)Seg · DEFACTO-12kCom-F1: 0.617 -> 0.579Hurt
add noise-sensitive branch (NSB) to Seg+Clf (ablation)Seg+Clf · DEFACTO-12kCom-F1: 0.579 -> 0.619Improved
add edge-supervised branch (ESB) to Seg+Clf (ablation)Seg+Clf · DEFACTO-12kCom-F1: 0.579 -> 0.626Improved
add Sobel layer to ESB (ablation)Seg+Clf+E · DEFACTO-12kcopy-move pixel F1: 0.382 -> 0.405Improved
use ESB instead of GSR-Net edge branch (ablation)Seg+Clf · DEFACTO-12kCom-F1: 0.613 -> 0.626Improved
fuse ESB and NSB with dual attention instead of model averaging (ablation)MVSS-Net · DEFACTO-12kCom-F1: 0.622 -> 0.643Improved
combine ESB and NSB (full setup) versus ESB alone (ablation)MVSS-Net · DEFACTO-12kCom-F1: 0.626 -> 0.643Improved
full multi-view multi-scale MVSS-Net versus plain FCN, fixed thresholdMVSS-Net · 5 test sets (trained on CASIAv2)pixel-level mean F1: 0.232 -> 0.394Improved
MVSS-Net versus FCN on inpainting subsets: optimal threshold meanMVSS-Net · 5 test sets (trained on CASIAv2)pixel-level mean F1: 0.562 -> 0.718Improved
MVSS-Net versus HP-FCN on inpainting subset of NIST16MVSS-Net · NIST16 inpaintingpixel-level F1: 0.284 -> 0.565Improved

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