A pipeline that combines geometric-mean Grad-CAM maps with pre-trained segmentation masks to locate manipulations without pixel labels, tested on a small biased subset of CASIA2.0.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Weakly-supervised Localization of Manipulated Image Regions Using Multi-resolution Learned Features
A pipeline that combines geometric-mean Grad-CAM maps with pre-trained segmentation masks to locate manipulations without pixel labels, tested on a small biased subset of CASIA2.0.