Research map / Image generation / Editing and style transfer
SAGI: Semantically Aligned and Uncertainty Guided AI Image Inpainting
Paschalis Giakoumoglou, Dimitrios Karageorgiou, Symeon Papadopoulos, Panagiotis C. Petrantonakis
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 |
|---|---|---|---|
| generate prompts with an LLM from object labels and captions (SAOR) instead of object labels only (ablation) | inpainting pipelines (avg of 5) · SAGI generation quality | Aesthetic Score: 5.56 -> 5.74 | Improved |
| LLM-generated prompts instead of labels and captions on technical quality metrics (ablation) | inpainting pipelines (avg of 5) · SAGI generation quality | QAlign Quality: 3.98 -> 3.96 | No clear effect |
| add UGDA uncertainty-guided filtering with GPT-4o (ablation) | SAGI inpainted images · human study | human accuracy at spotting inpainting: 54.9 -> 35.2 | Improved |
| split by existing quality/aesthetic metrics instead of UGDA to find deceiving images (ablation) | realism assessment · human study | accuracy gap between groups: +0.4 (AS) vs +38.5 (UGDA) | No clear effect |
| retrain forensic detectors on SAGI-D | TruFor · SAGI-D in-domain | Mean IoU: 12.2 -> 64.7 | Improved |
| retrain CAT-Net on SAGI-D, tested out-of-domain | CAT-Net · SAGI-D out-of-domain | Mean IoU: 20.9 -> 23.0 | No clear effect |
| train detectors on SAGI-D instead of TGIF | 4 forensic detectors · TGIF test set | detection AUC: 74.3 -> 78.3 | Improved |
| train detectors on TGIF and test on TGIF | 4 forensic detectors · TGIF test set | detection AUC: 81.2 -> 74.3 | Hurt |
| retrain detectors on SAGI-D vs TGIF, tested on SAGI-D | 4 forensic detectors · SAGI-D test set | detection AUC: 73.6 (TGIF) -> 93.1 (SAGI-D) | Improved |
Closest papers by meaning
- PrefPaint: Aligning Image Inpainting Diffusion Model with Human Preference (NeurIPS 2024 · similarity 0.87)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · similarity 0.87)
- Yuan: Yielding Unblemished Aesthetics Through a Unified Network for Visual Imperfections Removal in Generated Images (AAAI 2025 · similarity 0.87)
- HD-Painter: High-Resolution and Prompt-Faithful Text-Guided Image Inpainting with Diffusion Models (ICLR 2025 · similarity 0.87)
- Paint by Inpaint: Learning to Add Image Objects by Removing Them First (CVPR 2025 · similarity 0.86)
- MTADiffusion: Mask Text Alignment Diffusion Model for Object Inpainting (CVPR 2025 · 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.
