Research map / Image generation / Autoregressive and latent generation
DeepForgeSeal: Latent Space-Driven Semi-Fragile Watermarking for Deepfake Detection Using Adversarial Reinforcement Learning
Tharindu Fernando, Clinton Fookes, Sridha Sridharan
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 learnable directional embedding in the latent sphere (vs naive latent embedding) (ablation) | DeepForgeSeal · CelebA-HQ validation | BRA malicious: 0.42 -> 0.12 | Improved |
| add learnable directional embedding (vs naive latent embedding) (ablation) | DeepForgeSeal · CelebA-HQ validation | BRA benign: 0.78 -> 0.99 | Improved |
| add learnable directions (vs fixed directions) (ablation) | DeepForgeSeal · CelebA-HQ validation | BRA benign: 0.86 -> 0.99 | Improved |
| add learnable directional embedding (vs naive latent embedding) (ablation) | DeepForgeSeal · CelebA-HQ validation | PSNR: 46.23 -> 48.12 | Improved |
| add the attacker agent (ablation) | DeepForgeSeal · CelebA-HQ validation | BRA malicious: 0.59 -> 0.12 | Improved |
| add the attacker agent (ablation) | DeepForgeSeal · CelebA-HQ validation | PSNR: 32.11 -> 48.12 | Improved |
| use a learned combinatorial attack with adaptive curriculum (vs one fixed attack type) (ablation) | DeepForgeSeal · CelebA-HQ validation | BRA benign: 0.72 -> 0.99 | Improved |
| adapt the attack curriculum to the watermarker (vs predefined curriculum) (ablation) | DeepForgeSeal · CelebA-HQ validation | BRA malicious: 0.28 -> 0.12 | Improved |
| let a learnable agent compose attack sequences (vs random combinations) (ablation) | DeepForgeSeal · CelebA-HQ validation | BRA malicious: 0.31 -> 0.12 | Improved |
| add proximity and curiosity rewards (ablation) | DeepForgeSeal · CelebA-HQ validation | BRA malicious: 0.35 -> 0.12 | Improved |
| add curiosity reward to proximity reward (ablation) | DeepForgeSeal · CelebA-HQ validation | BRA malicious: 0.21 -> 0.12 | Improved |
| add proximity reward to curiosity reward (ablation) | DeepForgeSeal · CelebA-HQ validation | BRA malicious: 0.24 -> 0.12 | Improved |
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.
- Omniguard — OmniGuard: Hybrid Manipulation Localization via Augmented Versatile Deep Image Watermarking (CVPR 2025) · better on 1 of 1 shared result
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
- Robustness of AI-Image Detectors: Fundamental Limits and Practical Attacks (ICLR 2024 · similarity 0.88)
- Black-Box Forgery Attacks on Semantic Watermarks for Diffusion Models (CVPR 2025 · similarity 0.87)
- SleeperMark: Towards Robust Watermark against Fine-Tuning Text-to-image Diffusion Models (CVPR 2025 · similarity 0.86)
- Veritas: Generalizable Deepfake Detection via Pattern-Aware Reasoning (ICLR 2026 · similarity 0.86)
- RAW: A Robust and Agile Plug-and-Play Watermark Framework for AI-Generated Images with Provable Guarantees (NeurIPS 2024 · similarity 0.86)
- Lightweight-Mark: Rethinking Deep Learning-Based Watermarking (ICML 2025 · similarity 0.85)
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.
