Research map / Image generation / Fast diffusion sampling
ColorFM: An Optimization-to-Learning Framework for Color Transfer via Flow Matching
Yuhang He, Kai Zhang, Xiaoming Li, Du Chen, Jian 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 |
|---|---|---|---|
| increase subdivision depth Dmax from 0 to 3 (ablation) | hierarchical color coupling · ColorFM-O coupling ablation | Lipschitz: 5.811 -> 2.937 | Improved |
| increase Dmax from 3 to 5 (ablation) | hierarchical color coupling · ColorFM-O coupling ablation | Style similarity / time: 0.779 -> 0.773; 19.26 -> 24.37 s | Hurt |
| use hierarchical coupling (Dmax=3) instead of 2-Rectified Flow (ablation) | ColorFM-O · ColorFM-O coupling ablation | Style similarity: 0.738 -> 0.779 | Improved |
| use hierarchical coupling instead of Mini-Batch OT (ablation) | ColorFM-O · ColorFM-O coupling ablation | Time (s): 51.48 -> 19.26 | Improved |
| add explicit semantic alignment (ablation) | ColorFM-O · color transfer | qualitative | Improved |
| use bidirectional instead of direct transport (ablation) | ColorFM-L · ColorFM-L component ablation | Distance to Ideal: 0.339 -> 0.326 | Improved |
| add fixed state (ablation) | ColorFM-L · ColorFM-L component ablation | Distance to Ideal: 0.326 -> 0.334 | Hurt |
| add implicit state with concat fusion (ablation) | ColorFM-L · ColorFM-L component ablation | Distance to Ideal: 0.326 -> 0.331 | No clear effect |
| use cross-attention instead of concat fusion for implicit state (ablation) | ColorFM-L · ColorFM-L component ablation | Distance to Ideal: 0.331 -> 0.320 | Improved |
| train ColorFM-L on ColorFM-O pseudo pairs instead of instance optimization | ColorFM-O · color transfer (512x512) | Distance to Ideal: 0.338 -> 0.320 | Improved |
| degrade semantic masks by 20% misclassification (ablation) | ColorFM-O · semantic mask robustness | Style similarity: 0.745 -> 0.731 | No clear effect |
| degrade semantic masks by 60% misclassification (ablation) | ColorFM-O · semantic mask robustness | Style similarity: 0.745 -> 0.695 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SA-LUT — SA-LUT: Spatial Adaptive 4D Look-Up Table for Photorealistic Style Transfer (ICCV 2025) · better on 2 of 2 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
- Nonlinear Color Transfer via Learnable Bezier Flows (CVPR 2026 · similarity 0.88)
- Color Transfer with Modulated Flows (AAAI 2025 · similarity 0.87)
- Flow Straight to Reality: Perceptually Consistent Flow Matching for Efficient Image Restoration (ECCV 2026 · similarity 0.86)
- Balancing Frequencies and Pixels in Flow Matching (NeurIPS 2026 · similarity 0.86)
- Color Alignment in Diffusion (CVPR 2025 · similarity 0.85)
- Contextual Flow Matching: Adaptive Step Selection in Flow Models for Efficient Visual Generation (NeurIPS 2026 · similarity 0.85)
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.
