Research map / Label-efficient and robust learning / Self-supervised and contrastive
CanonCGT: Reference-Based Color Grading via Canonical Pivot Representation
Jinwon Ko, Keunsoo Ko, Chang-Su Kim
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 two-stage canonicalize-then-grade instead of one-stage (ablation) | CanonCGT · FiveK val. | PSNR: 29.43 -> 30.29 | Improved |
| use two-stage instead of one-stage (ablation) | CanonCGT · FiveK val. | Delta E_ab: 4.53 -> 4.25 | Improved |
| add self-supervised phase (DP-CGT) (ablation) | CanonCGT · Unsupervised test sets | PSNR: 19.17 -> 28.99 | Improved |
| add self-supervised phase (DP-CGT) (ablation) | CanonCGT · FiveK val. | PSNR: 30.29 -> 29.80 | Hurt |
| use Expert C canonical style instead of Warm-Contrast preset (ablation) | CanonCGT · FiveK val. | PSNR: 25.82 -> 30.29 | Improved |
| use Expert C canonical style instead of preset-mean (ablation) | CanonCGT · FiveK val. | PSNR: 29.80 -> 30.29 | Improved |
| increase query tokens from 9^3 to 17^3 (ablation) | CanonCGT · FiveK val. | PSNR: 29.87 -> 30.29 | Improved |
| increase query tokens from 17^3 to 33^3 (ablation) | CanonCGT · FiveK val. | PSNR: 30.29 -> 30.31 | No clear effect |
| increase channels from 32 to 64 (ablation) | CanonCGT · FiveK val. | PSNR: 29.74 -> 30.29 | Improved |
| increase channels from 64 to 96 (ablation) | CanonCGT · FiveK val. | PSNR: 30.29 -> 30.32 | No clear effect |
| increase attention heads from 2 to 4 (ablation) | CanonCGT · FiveK val. | PSNR: 29.68 -> 30.29 | Improved |
| increase attention heads from 4 to 8 (ablation) | CanonCGT · FiveK val. | PSNR: 30.29 -> 30.11 | Hurt |
Closest papers by meaning
- Video Color Grading via Look-Up Table Generation (ICCV 2025 · similarity 0.87)
- Eliminating Gradient Conflict in Reference-Based Line-Art Colorization (ECCV 2022 · similarity 0.86)
- AceTone: Bridging Words and Colors for Conditional Image Grading (CVPR 2026 · similarity 0.86)
- Towards In-Context Tone Style Transfer with A Large-Scale Triplet Dataset (ECCV 2026 · similarity 0.86)
- MangaNinja: Line Art Colorization with Precise Reference Following (CVPR 2025 · similarity 0.85)
- GLUT: 3D Gaussian Lookup Table for Continuous Color Transformation (NeurIPS 2026 · 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.
