Research map / Image generation / Editing and style transfer
Detecting Photoshopped Faces by Scripting Photoshop
Sheng-Yu Wang, Oliver Wang, Andrew Owens, Richard Zhang, Alexei A. Efros
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add augmentation (resizing, JPEG, color) to low-res classifier (ablation) | Low-res global classifier · Validation (random FAL) | accuracy (%): 97.0 -> 93.7 | Hurt |
| add augmentation to low-res classifier after Facebook post-processing (ablation) | Low-res global classifier · Artist test set, Facebook post-processing | accuracy (%): 57.0 -> 67.0 | Improved |
| increase input resolution from 400 to 700 | Global classifier with aug. · Validation (random FAL) | accuracy (%): 93.7 -> 97.1 | Improved |
| increase input resolution to 700 on Facebook post-processed images | Global classifier with aug. · Artist test set, Facebook post-processing | accuracy (%): 67.0 -> 55.0 | Hurt |
| add reconstruction loss to EPE and multiscale gradient losses (ablation) | Local flow prediction model · Val (Rand-FAL) | ΔPSNR (dB): +2.38 -> +2.69 | Improved |
| add reconstruction loss on top of MultiG for Artist-Liquify (ablation) | Local flow prediction model · Artist-Liquify | ΔPSNR (dB): -0.84 -> -0.72 | Improved |
| add multiscale gradient and reconstruction losses to EPE-only (ablation) | Local flow prediction model · Val (Rand-FAL) | EPE (lower better): 0.51 -> 0.52 | Mixed |
| train on random FAL, test on artist-made FAL | High-res global classifier · Validation vs Artist test | accuracy (%): 97.1 -> 90.0 | Hurt |
| test on generic Liquify instead of Face-Aware Liquify | Global classifier · Artist test set | accuracy (%): 90.0 -> 64.0 | Hurt |
| test on out-of-distribution warps (X2face flows) instead of FAL on faces | Global classifier · Generalization | accuracy (%): 93.7 -> 64.7 | Hurt |
| print and rescan images (rebroadcasting) | Global classifier · Robustness | accuracy (%): 94.2 -> 69.2 | Hurt |
Closest papers by meaning
- FaceForensics++: Learning to Detect Manipulated Facial Images (ICCV 2019 · similarity 0.87)
- Face Forgery Detection by 3D Decomposition and Composition Search (TPAMI 2023 · similarity 0.86)
- Face Forgery Detection by 3D Decomposition (CVPR 2021 · similarity 0.85)
- FakeShield: Explainable Image Forgery Detection and Localization via Multi-modal Large Language Models (ICLR 2025 · similarity 0.85)
- Edit Away and My Face Will not Stay: Personal Biometric Defense against Malicious Generative Editing (CVPR 2025 · similarity 0.83)
- Identity-Aware Vision-Language Model for Explainable Face Forgery Detection (AAAI 2026 · similarity 0.83)
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.
