Research map / Image generation / Image restoration
ZITS++: Image Inpainting by Improving the Incremental Transformer on Structural Priors
Chenjie Cao, Qiaole Dong, Yanwei Fu
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 ZITS++ instead of ZITS | image inpainting · Indoor 256x256 | FID: 15.93 -> 14.61 | Improved |
| use ZITS++ instead of LaMa | LaMa · Places2 256x256 | FID: 1.63 -> 1.19 | Improved |
| use ZITS++ instead of LaMa | LaMa · MatterPort3D 1024x1024 | FID: 14.04 -> 10.86 | Improved |
| add MPE (ablation) | FTR (LaMa-style) · Indoor 256x256 | FID: 16.97 -> 16.44 | Improved |
| add ZeroRA when incorporating TSR structure (ablation) | ZITS · Indoor 256x256 | PSNR: 25.28 -> 25.57 | Improved |
| add new priors (L-Edge and gradient) (ablation) | ZITS · Indoor 256x256 | FID: 15.93 -> 14.66 | Improved |
| add LKA-enhanced FTR (ablation) | ZITS++ without LKA · Indoor 256x256 | FID: 14.66 -> 14.61 | Improved |
| add MPE during high-resolution finetuning (ablation) | ZITS++ · Places2 512x512 | FID: 25.95 -> 25.64 | Improved |
| use LKA(21) instead of FFC upsampling/downsampling convs (ablation) | FTR · Places2 subset 256x256 | FID: 24.15 -> 23.31 | Improved |
| use LKA kernel 21 instead of 14 or 28 (ablation) | FTR · Places2 subset 256x256 | FID: 23.76 -> 23.31 (14 to 21); 23.56 (28) | Mixed |
| use CATS edges instead of canny edges as prior (ablation) | ZITS++ · Places2 subset 256x256 | FID: 21.68 -> 21.50 | Improved |
| add gradient prior (ablation) | baseline LaMa · Places2 subset 256x256 | PSNR: 25.09 -> 25.80 | Improved |
Later papers that compared with this method
2 later accepted papers report ZITS as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- High-Fidelity Image Inpainting with Multimodal Guided GAN Inversion (IJCV 2025 · better on 2 of 2 shared results)
- CoordFill: Efficient High-Resolution Image Inpainting via Parameterized Coordinate Querying (AAAI 2023 · better on 2 of 3 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
- Incremental Transformer Structure Enhanced Image Inpainting With Masking Positional Encoding (CVPR 2022 · similarity 0.92)
- SyFormer: Structure-Guided Synergism Transformer for Large-Portion Image Inpainting (AAAI 2024 · similarity 0.87)
- Unbiased Multi-Modality Guidance for Image Inpainting (ECCV 2022 · similarity 0.86)
- StructureFlow: Image Inpainting via Structure-Aware Appearance Flow (ICCV 2019 · similarity 0.86)
- Towards Enhanced Image Inpainting: Mitigating Unwanted Object Insertion and Preserving Color Consistency (CVPR 2025 · similarity 0.85)
- Paint by Inpaint: Learning to Add Image Objects by Removing Them First (CVPR 2025 · 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.
