Research map / Image generation / Autoregressive and latent generation
Detecting Compressed AI-Generated Images via Phase Spectrum Robustness
Kai Li, Wenqi Ren, Wei Wang, Xiaochun Cao
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 phase-harmonized cross-modal interaction (PHCI) to CLIP-ViT baseline (ablation) | CLIP-ViT baseline · GANs, quality-aware, 2-class | mean accuracy (%): 64.6 -> 74.7 | Improved |
| add multi-domain modulation adapter (MDMA) (ablation) | CLIP-ViT + PHCI · GANs, quality-aware, 2-class | mean accuracy (%): 74.7 -> 76.0 | Improved |
| add difficulty-aware consistency loss (DACL) (ablation) | CLIP-ViT + PHCI + MDMA · GANs, quality-aware, 2-class | mean accuracy (%): 76.0 -> 77.4 | Improved |
| add PHCI on specific generators (BigGAN, CycleGAN, StarGAN, GauGAN) vs baseline (ablation) | CLIP-ViT baseline · GANs, quality-aware, 2-class | accuracy on Att-GAN: Att-GAN 54.7 -> 58.5; StyleGAN2 60.8 -> 56.3 | Mixed |
| use CPTFormer instead of ODDN with 2-class training, compressed test images | AI-generated image detector · GANs, quality-aware | mean accuracy (%): 71.4 -> 77.4 | Improved |
| use CPTFormer instead of best baseline on diffusion models | AI-generated image detector · Diffusion models, quality-aware | mean accuracy (%): 61.4 -> 65.9 | Improved |
| use CPTFormer in random-quality compression setting for GANs | AI-generated image detector · GANs, quality-agnostic, 2-class | mean accuracy (%): 70.7 -> 76.3 | 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.
- ODDN — ODDN: Addressing Unpaired Data Challenges in Open-World Deepfake Detection on Online Social Networks (AAAI 2025) · better on 5 of 5 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
- Decision-Driven Orthogonal Learning with Complementary Feature Mining for Robust Synthetic Image Detection (AAAI 2026 · similarity 0.85)
- Any-Resolution AI-Generated Image Detection by Spectral Learning (CVPR 2025 · similarity 0.84)
- Compression of Structured Data with Autoencoders: Provable Benefit of Nonlinearities and Depth (ICML 2024 · similarity 0.83)
- Stay-Positive: A Case for Ignoring Real Image Features in Fake Image Detection (ICML 2025 · similarity 0.83)
- Training-free Detection of AI-generated images via Cropping Robustness (NeurIPS 2025 · similarity 0.82)
- Robustness-Guided Image Synthesis for Data-Free Quantization (AAAI 2024 · similarity 0.82)
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.
