Research map / Image generation / Editing and style transfer
TransFill: Reference-Guided Image Inpainting by Merging Multiple Color and Spatial Transformations
Yuqian Zhou, Connelly Barnes, Eli Shechtman, Sohrab Amirghodsi
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 TransFill instead of OPN | RealEstate10K All · RealEstate10K All | PSNR: 32.43 -> 37.58 | Improved |
| use TransFill instead of ProFill | RealEstate10K Small Set · RealEstate10K Small Set | PSNR: 30.95 -> 38.83 | Improved |
| use multiple depth-clustered homographies instead of one global homography (ablation) | TransFill · RealEstate10K | PSNR: 37.092 -> 37.576 | Improved |
| use depth clustering rather than random clustering (ablation) | TransFill · RealEstate10K | PSNR: 37.499 -> 37.576 | Improved |
| use depth clustering rather than spatial clustering (ablation) | TransFill · RealEstate10K | PSNR: 37.384 -> 37.576 | Improved |
| increase the proposal count from N=3 to N=5 (ablation) | TransFill · RealEstate10K | PSNR: 37.537 -> 37.576 | Improved |
| add color transformation to the spatial-only CST (ablation) | TransFill CST · RealEstate10K | PSNR: 36.717 -> 37.576 | Improved |
| add spatial transformation to the color-only CST (ablation) | TransFill CST · RealEstate10K | PSNR: 36.228 -> 37.576 | Improved |
| reverse color-then-spatial transformation to spatial-then-color (ablation) | TransFill CST · RealEstate10K | PSNR: 37.576 -> 37.566 | No clear effect |
| add CST to the pipeline (ablation) | TransFill · RealEstate10K | PSNR: 35.579 -> 37.576 | Improved |
| add SPF to the pipeline (ablation) | TransFill · RealEstate10K | PSNR: 36.710 -> 37.576 | Improved |
| add Poisson blending after APAP (ablation) | APAP · RealEstate10K Small Set | PSNR: 31.94dB -> 32.56dB | Mixed |
Later papers that compared with this method
1 later accepted paper reports TransFill as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- LeftRefill: Filling Right Canvas based on Left Reference through Generalized Text-to-Image Diffusion Model (CVPR 2024 · better on 0 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Diverse Image Inpainting with Normalizing Flow (ECCV 2022 · similarity 0.86)
- Contextual-based Image Inpainting: Infer, Match, and Translate (ECCV 2018 · similarity 0.86)
- HomoGen: Enhanced Video Inpainting via Homography Propagation and Diffusion (CVPR 2025 · similarity 0.85)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · similarity 0.85)
- TurboFill: Adapting Few-step Text-to-image Model for Fast Image Inpainting (CVPR 2025 · similarity 0.85)
- Reconstructing the Image Stitching Pipeline: Integrating Fusion and Rectangling into a Unified Inpainting Model (NeurIPS 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.
