Research map / Image generation / GANs
RepMix: Representation Mixing for Robust Attribution of Synthesized Images
Tu Bui, Ning Yu, John Collomosse
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 RepMix layer (ablation) | RepMix · Attribution88 subset (2 train semantics) | attribution accuracy: 0.7188 -> 0.7400 | Improved |
| add compound loss (ablation) | RepMix · Attribution88 subset (2 train semantics) | attribution accuracy: 0.7204 -> 0.7400 | Improved |
| add compound loss (ablation) | RepMix · Attribution88 subset (2 train semantics) | detection accuracy: 0.9364 -> 0.9426 | Improved |
| add ImageNet-C perturbation augmentation (ablation) | RepMix · Attribution88 subset (2 train semantics) | attribution accuracy: 0.2762 -> 0.7400 | Improved |
| use ResNet50 instead of AlexNet backbone (ablation) | RepMix · Attribution88 subset (2 train semantics) | attribution accuracy: 0.5280 -> 0.7400 | Improved |
| use ResNet50 instead of VGG16 backbone (ablation) | RepMix · Attribution88 subset (2 train semantics) | attribution accuracy: 0.7150 -> 0.7400 | Improved |
| use RepMix instead of Yu et al. approach | RepMix · Attribution88 | attribution accuracy: 0.6784 -> 0.8207 | Improved |
| use RepMix instead of Yu et al. under I-FGSM attack | RepMix · Attribution88 | accuracy at eps=32/255: 22.49% -> 46.35% | Improved |
| mix images at pixel level (MixUp) instead of at a later feature layer (ablation) | RepMix · Attribution88 | attribution accuracy | Hurt |
| increase number of mixed samples to 4 (ablation) | RepMix · Attribution88 | attribution accuracy: 1% improvement at 4 mixing samples at most | No clear effect |
| increase number of seen semantics from 2 to 11 (ablation) | RepMix · Attribution88 | attribution accuracy: 59% -> 90% | Improved |
Closest papers by meaning
- Attributing Fake Images to GANs: Learning and Analyzing GAN Fingerprints (ICCV 2019 · similarity 0.92)
- Towards Discovery and Attribution of Open-World GAN Generated Images (ICCV 2021 · similarity 0.89)
- Artificial Fingerprinting for Generative Models: Rooting Deepfake Attribution in Training Data (ICCV 2021 · similarity 0.88)
- Attributing Image Generative Models using Latent Fingerprints (ICML 2023 · similarity 0.87)
- One for All: Synthesis-Free Fingerprint Learning for Attribution of In-the-Wild Synthetic Images (AAAI 2026 · similarity 0.87)
- AdaptiveMix: Improving GAN Training via Feature Space Shrinkage (CVPR 2023 · similarity 0.86)
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.
