Research map / Graph learning / Knowledge and temporal graphs
LGA-Net: Learning Local and Global Affinities for Sparse Scribble based Image Colorization
Hongjin Lyu, Bo Li, Paul L. Rosin, Yu-Kun Lai
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 global points/affinities (ablation) | LGA-Net without global affinities · Dmanual | PSNR: 21.89 -> 26.07 | Improved |
| add global points/affinities (ablation) | LGA-Net without global affinities · Dmanual | LPIPS (x10^-2): 15.38 -> 10.49 | Improved |
| add local points/affinities (ablation) | LGA-Net with global only · Dmanual | PSNR: 24.40 -> 26.07 | Improved |
| add local points/affinities (ablation) | LGA-Net with global only · Dauto | PSNR: 28.82 -> 29.89 | Improved |
| add global affinities (ablation) | LGA-Net without global affinities · Dauto | PSNR: 28.16 -> 29.89 | Improved |
| add Non-local Block (ablation) | LGA-Net without NLB · Dmanual | PSNR: 25.82 -> 26.07 | Improved |
| add Non-local Block (ablation) | LGA-Net without NLB · Dauto | LPIPS (x10^-2): 10.10 -> 9.62 | Improved |
| use dot product affinity instead of concatenation in NLB (ablation) | LGA-Net · Dmanual | PSNR: 25.77 -> 26.07 | Improved |
| use dot product instead of Gaussian in NLB (ablation) | LGA-Net · Dmanual | LPIPS (x10^-2): 10.98 -> 10.49 | Improved |
| increase global step GS (sparser global points) (ablation) | LGA-Net · Dmanual | LPIPS / MS-SSIM | Hurt |
| train on 4K images instead of ImageNet 1.3M | iColoriT · Dmanual | PSNR: 25.92 -> 21.32 | Hurt |
| compare LGA-Net (4K training) with UniColor (1.3M training) | LGA-Net vs UC1M · Dmanual | PSNR: 25.36 -> 26.07 | Improved |
Closest papers by meaning
- Semantic-Sparse Colorization Network for Deep Exemplar-Based Colorization (ECCV 2022 · similarity 0.89)
- Enabling Region-Specific Control via Lassos in Point-Based Colorization (AAAI 2025 · similarity 0.87)
- SCSNet: An Efficient Paradigm for Learning Simultaneously Image Colorization and Super-resolution (AAAI 2022 · similarity 0.87)
- Colorful Image Colorization (ECCV 2016 · similarity 0.86)
- Control Color: Multimodal Diffusion-Based Interactive Image Colorization (IJCV 2025 · similarity 0.86)
- Learning Diverse Image Colorization (CVPR 2017 · similarity 0.86)
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.
