Research map / Architectures and efficiency / Attention and transformers
Continuously Masked Transformer for Image Inpainting
Keunsoo Ko, Chang-Su Kim
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 |
|---|---|---|---|
| use overlapping tokens instead of ordinary tokens (ablation) | CMT with standard self-attention · Places2 H2I (0.3, 0.4] | FID: 22.39 -> 19.83 | Improved |
| add masked self-attention and mask update (continuous) with ordinary tokens (ablation) | CMT · Places2 H2I (0.3, 0.4] | FID: 22.39 -> 19.62 | Improved |
| use continuous mask instead of binary mask in MSAU (overlapping tokens) (ablation) | CMT · Places2 H2I (0.3, 0.4] | FID: 18.83 -> 15.69 | Improved |
| use continuous mask instead of binary mask in MSAU (ordinary tokens) (ablation) | CMT · Places2 H2I (0.3, 0.4] | PSNR: 22.51 -> 22.83 | Improved |
| add overlapping tokens to MSAU with continuous mask (ablation) | CMT · Places2 H2I (0.3, 0.4] | PSNR: 22.83 -> 23.22 | Improved |
| add refinement network (ablation) | CMT · Places2 H2I (0.3, 0.4] | FID: 15.69 -> 14.69 | Improved |
| replace MAT with CMT | image inpainting · Places2 H2I (0.3, 0.4] | PSNR: 22.20 -> 23.56 | Improved |
| replace MAT with CMT on CelebA-HQ with large holes | image inpainting · CelebA-HQ H2I (0.4, 0.6] | FID: 4.60 -> 5.23 | Hurt |
| replace BAT with CMT on out-of-domain texture images | image inpainting · DTD H2I (0.3, 0.4] | PSNR: 22.62 -> 24.05 | Improved |
Closest papers by meaning
- MAT: Mask-Aware Transformer for Large Hole Image Inpainting (CVPR 2022 · similarity 0.90)
- Token Painter: Training-Free Text-Guided Image Inpainting via Mask Autoregressive Models (AAAI 2026 · similarity 0.87)
- Incremental Transformer Structure Enhanced Image Inpainting With Masking Positional Encoding (CVPR 2022 · similarity 0.87)
- Reduce Information Loss in Transformers for Pluralistic Image Inpainting (CVPR 2022 · similarity 0.86)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · similarity 0.85)
- Don't Look into the Dark: Latent Codes for Pluralistic Image Inpainting (CVPR 2024 · similarity 0.84)
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.
