Research map / Image generation / Autoregressive and latent generation
Perceptual Artifacts Localization for Image Synthesis Tasks
Lingzhi Zhang, Zhengjie Xu, Connelly Barnes, Yuqian Zhou, Qing Liu, He Zhang, Sohrab Amirghodsi, Zhe Lin, Eli Shechtman, Jianbo Shi
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 |
|---|---|---|---|
| train one unified artifact segmentation model on all tasks instead of per-task specialists (ablation) | Swin-T + UperNet artifact detector · 10 synthesis tasks | mIoU: PSR 21.33 -> 5.10; StyleGAN2 37.85 -> 38.53; AnyRes 9.15 -> 34.74 | Mixed |
| fine-tune the pretrained detector with about 10 labeled images on an unseen generator (ablation) | artifact detector · Versatile Diffusion | mIoU: 5.92 -> 22.18 | Improved |
| fine-tune the pretrained detector with about 10 labeled images on an unseen generator (ablation) | artifact detector · StyleGAN3 | mIoU: 46.45 -> 40.81 | Hurt |
| fine-tune the pretrained detector with about 10 labeled images on an unseen generator (ablation) | artifact detector · DiT | mIoU: 16.46 -> 31.76 | Improved |
| train on the new multi-task artifact dataset versus prior detectors | artifact segmentation · StyleGAN3 (unseen) | mIoU: 11.43 -> 46.45 | Improved |
| use PAR as no-reference quality metric versus MUSIQ and HyperIQA | image quality ranking · Stable Diffusion pairs | agreement with human preference: 48.45% -> 63.92% | Improved |
| refine detected artifact regions with zoom-in inpainting using SD inpainting | Stable Diffusion text2image · 5,000 SD images | FID-CLIP: 15.98 -> 13.28 | Improved |
| gradually zoom in the inpainting input context (ablation) | diffusion inpainting of faces and hands · 100 images | Perceptual Artifacts Ratio | Improved |
| use CNNgenerates real/fake classifier with Grad-CAM as artifact localizer | existing detector baseline · 10 synthesis tasks | mIoU: StyleGAN2 4.38 vs 38.53 for unified model | Hurt |
Closest papers by meaning
- Perceptual Artifacts Localization for Inpainting (ECCV 2022 · similarity 0.91)
- SALART-VQA: Diagnosing Whether VLMs Understand Salient Artifacts in Generated Images (NeurIPS 2026 · similarity 0.87)
- Can We Model the Artifacts Explicitly? Disentangle Artifacts via Pairwise Edit Relations for Image Manipulation Localization (NeurIPS 2026 · similarity 0.86)
- Puzzle Similarity: A Perceptually-guided Cross-Reference Metric for Artifact Detection in 3D Scene Reconstructions (ICCV 2025 · similarity 0.86)
- PixPerfect: Seamless Latent Diffusion Local Editing with Discriminative Pixel-Space Refinement (NeurIPS 2025 · similarity 0.85)
- Refine-by-Align: Reference-Guided Artifacts Refinement through Semantic Alignment (ICLR 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.
