Research map / Image generation / Autoregressive and latent generation
Incremental Transformer Structure Enhanced Image Inpainting With Masking Positional Encoding
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 transformer-restored structure features and masked positional encoding to the pretrained FTR inpainting backbone | FTR, evaluated as ZITS · Indoor (256×256), mixed masks | PSNR: 25.20 -> 25.57 | Helped |
| Add transformer-restored structure features and masked positional encoding to the pretrained FTR inpainting backbone | FTR, evaluated as ZITS · Places2 (256×256), mixed masks | LPIPS: 0.155 -> 0.108 | Helped |
| Add MPE (ablation) | FTR on Indoor · Indoor | PSNR: 25.20 -> 25.31 | Helped |
| Add structural information without ZeroRA (ablation) | FTR on Indoor · Indoor | PSNR: 25.20 -> 25.28 | No clear effect |
| Add ZeroRA for incremental fusion of structural features (ablation) | FTR on Indoor · Indoor | PSNR: 25.28 -> 25.57 | Helped |
| Add gated convolutions in the structure feature encoder (ablation) | FTR + SFE + MPE + ReZero · Indoor | PSNR: 25.51 -> 25.57 | Helped |
| Add MPE (ablation) | ZITS during 512×512 Places2 finetuning · Places2 (512×512) | FID: 26.29 -> 26.08 | Helped |
| Replace one standard attention module with axial attention (ablation) | TSR · Indoor, inference | FPS: 6.41 -> 7.89 | Helped |
| Replace one standard attention module with axial attention (ablation) | TSR · Indoor, inference | GPU memory: 14845 -> 10547 | Helped |
| Add axial attention while retaining the same metric setting (ablation) | TSR · Indoor edge and line reconstruction | Average F1: 46.26 -> 46.26 | No clear effect |
| Add relative position encoding to axial attention (ablation) | TSR · Indoor edge and line reconstruction | Average F1: 46.26 -> 46.72 | Helped |
| Multiply line logits by 4 before sigmoid (ablation) | TSR · Indoor and Places2 line reconstruction | Recall | Mixed |
Closest papers by meaning
- MAT: Mask-Aware Transformer for Large Hole Image Inpainting (CVPR 2022 · similarity 0.90)
- SyFormer: Structure-Guided Synergism Transformer for Large-Portion Image Inpainting (AAAI 2024 · similarity 0.90)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · similarity 0.87)
- IMFine: 3D Inpainting via Geometry-guided Multi-view Refinement (CVPR 2025 · similarity 0.86)
- Paint by Inpaint: Learning to Add Image Objects by Removing Them First (CVPR 2025 · similarity 0.86)
- Reconstructing the Image Stitching Pipeline: Integrating Fusion and Rectangling into a Unified Inpainting Model (NeurIPS 2024 · 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.