Research map / Architectures and efficiency / Attention and transformers
MAT: Mask-Aware Transformer for Large Hole Image Inpainting
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 |
|---|---|---|---|
| Use transformer blocks instead of convolution blocks (ablation) | MAT · Places, 256×256 ablation validation | FID: 6.21 -> 5.97 | Helped |
| Use the adjusted transformer block instead of the original block (ablation) | MAT · Places, 256×256 ablation validation | FID: 6.36 -> 5.97 | Helped |
| Use multi-head contextual attention instead of conventional attention (ablation) | MAT · Places, 256×256 ablation validation | FID: 6.08 -> 5.97 | Helped |
| Use multi-head contextual attention instead of conventional attention (ablation) | MAT · Places, 256×256 ablation validation | U-IDS (%): 29.19 -> 29.23 | No clear effect |
| Add the noise style manipulation module (ablation) | MAT · Places, 256×256 ablation validation | P-IDS (%): 11.88 -> 13.17 | Helped |
| Add the noise style manipulation module (ablation) | MAT · Places, 256×256 ablation validation | U-IDS (%): 27.94 -> 29.23 | Helped |
| Generate the first-stage prediction at high resolution rather than limiting it to 64×64 (ablation) | MAT · Places, 256×256 ablation validation | FID: 6.32 -> 5.97 | Helped |
| Use stacked convolution tokenization instead of linear projection (ablation) | MAT · Places, 256×256 ablation validation | FID: 10.54 -> 5.97 | Helped |
| Increase feature dimension from 90 to 180 (ablation) | MAT · Places, 256×256 ablation validation | FID: 6.28 -> 5.97 | Helped |
| Use more transformer blocks, {2, 3, 4, 3, 2} instead of {1, 1, 2, 1, 1} (ablation) | MAT · Places, 256×256 ablation validation | FID: 6.18 -> 5.97 | Helped |
| Use the configured larger window sizes instead of window size 8 throughout (ablation) | MAT · Places, 256×256 ablation validation | FID: 6.09 -> 5.97 | Helped |
| Use MAT rather than CoModGAN | Places365-Standard · Places365-Standard, 512×512, large mask | FID: 2.92 -> 1.96 | Helped |
Closest papers by meaning
- SyFormer: Structure-Guided Synergism Transformer for Large-Portion Image Inpainting (AAAI 2024 · similarity 0.88)
- IMFine: 3D Inpainting via Geometry-guided Multi-view Refinement (CVPR 2025 · similarity 0.87)
- WaveFormer: Wavelet Transformer for Noise-Robust Video Inpainting (AAAI 2024 · similarity 0.86)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · similarity 0.86)
- Reconstructing the Image Stitching Pipeline: Integrating Fusion and Rectangling into a Unified Inpainting Model (NeurIPS 2024 · similarity 0.86)
- InpDiffusion: Image Inpainting Localization via Conditional Diffusion Models (AAAI 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.