Research map / Label-efficient and robust learning / Self-supervised and contrastive
SSH: A Self-Supervised Framework for Image Harmonization
Yifan Jiang, He Zhang, Jianming Zhang, Yilin Wang, Zhe Lin, Kalyan Sunkavalli, Simon Chen, Sohrab Amirghodsi, Sarah Kong, Zhangyang Wang
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 |
|---|---|---|---|
| add multi-crop content augmentation (vs single crop) (ablation) | SSH · RealHM | PSNR: 24.80 -> 27.91 | Improved |
| add multi-crop content augmentation (vs single crop) (ablation) | SSH · RealHM | LPIPS: 0.069 -> 0.039 | Improved |
| use 3D LUT appearance augmentation instead of color transfer (ablation) | SSH · RealHM | PSNR: 27.01 -> 27.91 | Improved |
| use 3D LUT appearance augmentation instead of color transfer (ablation) | SSH · RealHM | MSE: 311.96 -> 206.85 | Improved |
| use 3D LUT appearance augmentation instead of random saturation change (ablation) | SSH · RealHM | PSNR: 26.68 -> 27.91 | Improved |
| add reconstruction loss (ablation) | SSH · RealHM | PSNR: 24.05 -> 27.91 | Improved |
| add reconstruction loss (ablation) | SSH · RealHM | MSE: 475.45 -> 206.85 | Improved |
| add disentanglement loss (ablation) | SSH · RealHM | PSNR: 24.94 -> 27.91 | Improved |
| add disentanglement loss (ablation) | SSH · RealHM | LPIPS: 0.0643 -> 0.039 | Improved |
| use self-supervised SSH instead of mask-supervised DoveNet | DoveNet · RealHM | PSNR: 27.41 -> 27.91 | Improved |
| use self-supervised SSH instead of DoveNet in human study | DoveNet · RealHM human study | visual quality score: 1.256 -> 2.295 | Improved |
| compare direct composite to DIH output | DC · RealHM | PSNR: 25.91 -> 23.96 | Hurt |
Closest papers by meaning
- Training-and-Prompt-Free General Painterly Harmonization via Zero-Shot Disentenglement on Style and Content References (AAAI 2025 · similarity 0.87)
- Relightful Harmonization: Lighting-aware Portrait Background Replacement (CVPR 2024 · similarity 0.87)
- Progressive Painterly Image Harmonization from Low-Level Styles to High-Level Styles (AAAI 2024 · similarity 0.87)
- FRIH: Fine-Grained Region-Aware Image Harmonization (AAAI 2024 · similarity 0.87)
- Painterly Image Harmonization by Learning from Painterly Objects (AAAI 2024 · similarity 0.85)
- Video Harmonization with Triplet Spatio-Temporal Variation Patterns (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.
