Research map / Architectures and efficiency / Attention and transformers
Eliminating Gradient Conflict in Reference-Based Line-Art Colorization
Zekun Li, Zhengyang Geng, Zhao Kang, Wenyu Chen, Yibo Yang
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 stop-gradient to the attention (ablation) | SGA · anime | FID: 36.34 -> 29.65 | Improved |
| add stop-gradient to the attention (ablation) | SGA · cat | FID: 40.73 -> 34.35 | Improved |
| add stop-gradient to the attention (ablation) | SGA · dog | FID: 72.34 -> 54.76 | Improved |
| add stop-gradient to the attention (ablation) | SGA · wild | FID: 19.90 -> 15.19 | Improved |
| add stop-gradient to the attention (ablation) | SGA · anime | SSIM: 0.876 -> 0.912 | Improved |
| add double normalization instead of classic softmax (ablation) | SGA · anime | FID: 33.42 -> 29.65 | Improved |
| add double normalization instead of classic softmax (ablation) | SGA · anime | SSIM: 0.861 -> 0.912 | Improved |
| add double normalization instead of classic softmax (ablation) | SGA · cat | FID: 34.42 -> 34.35 | No clear effect |
| add self-SGA (ablation) | SGA · anime | FID: 31.56 -> 29.65 | Improved |
| add self-SGA (ablation) | SGA · anime | SSIM: 0.917 -> 0.912 | Hurt |
| add self-SGA (ablation) | SGA · cat | FID: 34.26 -> 34.35 | No clear effect |
| add stop-gradient to SCFT attention (ablation) | SCFT · anime | FID: 44.65 -> 38.20 | Improved |
Closest papers by meaning
- MangaNinja: Line Art Colorization with Precise Reference Following (CVPR 2025 · similarity 0.88)
- CanonCGT: Reference-Based Color Grading via Canonical Pivot Representation (CVPR 2026 · similarity 0.86)
- Image Referenced Sketch Colorization Based on Animation Creation Workflow (CVPR 2025 · similarity 0.86)
- Tag2Pix: Line Art Colorization Using Text Tag With SECat and Changing Loss (ICCV 2019 · similarity 0.86)
- Reference-Based Sketch Image Colorization Using Augmented-Self Reference and Dense Semantic Correspondence (CVPR 2020 · similarity 0.86)
- Learning Inclusion Matching for Animation Paint Bucket Colorization (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.
