Research map / Image generation / Image restoration
Beyond Fixed Luminance: Towards Panchromatic and Orthochromatic Image Colorization
Swarnim Maheshwari, Syed Imam Ali, Vineeth N Balasubramanian
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 |
|---|---|---|---|
| fine-tune FLUX.2-klein-4B with LoRA for colorization (vs no fine-tuning) (ablation) | Mixed-FT · COCO ortho structural fidelity | SSIM: 0.360 -> 0.811 | Improved |
| fine-tune on ImageNet colorization (vs no fine-tuning) (ablation) | Mixed-FT · ImageNet ortho structural fidelity | SSIM: 0.373 -> 0.842 | Improved |
| add orthochromatic data to training (Mixed vs Normal-FT) on ortho inputs (ablation) | Mixed-FT · COCO ortho structural fidelity | SSIM: 0.828 -> 0.811 | Hurt |
| add orthochromatic data to training (Mixed vs Normal-FT) on ortho inputs (ablation) | Mixed-FT · ImageNet ortho structural fidelity | L1: 0.050 -> 0.061 | Hurt |
| add orthochromatic data to training (Mixed vs Normal-FT) on pan inputs (ablation) | Mixed-FT · COCO pan structural fidelity | SSIM: 0.838 -> 0.841 | No clear effect |
| use luminance-agnostic Flux model instead of DDColor under ortho input | Ours · ImageNet ortho | FID: 8.74 -> 7.91 | Improved |
| use luminance-agnostic Flux model instead of DDColor under ortho input | Ours · COCO ortho | FID-DINO: 47.25 -> 43.32 | Improved |
| use luminance-agnostic Flux model instead of DDColor under ortho input | Ours · COCO ortho | FID: 10.15 -> 10.67 | Hurt |
| use luminance-agnostic Flux model instead of DDColor under pan input | Ours · ImageNet pan | FID: 5.51 -> 7.03 | Hurt |
| switch input from pan to ortho for DDColor | DDColor · ImageNet | FID: 5.51 -> 8.74 | Hurt |
| switch input from pan to ortho for Ours | Ours · ImageNet | FID: 7.03 -> 7.91 | Hurt |
| switch input from pan to ortho for BigColor | BigColor · Multi-Instance | FID-DINO: 51.48 -> 118.92 | Hurt |
Closest papers by meaning
- Recolour What Matters: Region-Aware Colour Editing via Token-Level Diffusion (ECCV 2026 · similarity 0.89)
- Automatic Controllable Colorization via Imagination (CVPR 2024 · similarity 0.89)
- Structural Consistency and Controllability for Diverse Colorization (ECCV 2018 · similarity 0.88)
- Control Color: Multimodal Diffusion-Based Interactive Image Colorization (IJCV 2025 · similarity 0.88)
- ChromaFusionNet (CFNet): Natural Fusion of Fine-Grained Color Editing (AAAI 2024 · similarity 0.87)
- Enabling Region-Specific Control via Lassos in Point-Based Colorization (AAAI 2025 · 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.
