Research map / Image generation / Image restoration
Cross-Camera Convolutional Color Constancy
Mahmoud Afifi, Jonathan T. Barron, Chloe LeGendre, Yun-Ta Tsai, Francois Bleibel
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 the gradient intensity histogram as input (ablation) | C5 · Cube+ ablation (7 encoders, trained with augmentation) | mean angular error: 2.30 -> 1.87 | Improved |
| add extra u,v coordinate channels (CoordConv) to histograms (ablation) | C5 · Cube+ ablation (7 encoders, trained with augmentation) | mean angular error: 2.03 -> 1.87 | Improved |
| add smoothness regularization on generated filters and bias (ablation) | C5 · Cube+ ablation (7 encoders, trained with augmentation) | mean angular error: 2.07 -> 1.87 | Improved |
| increase smoothness strength to lambda_B=0.25, lambda_F=1.85 (ablation) | C5 · Cube+ ablation (7 encoders, trained with augmentation) | mean angular error: 1.87 -> 2.23 | Hurt |
| increase batch size during training from 16 to 64 (ablation) | C5 · Cube+ ablation (7 encoders, trained with augmentation) | mean angular error: 1.93 -> 1.87 | Improved |
| add cross-camera data augmentation (ablation) | C5 · Cube+ ablation | mean angular error: 2.10 -> 1.87 | Improved |
| add cross-camera data augmentation (ablation) | C5 · INTEL-TAU ablation | mean angular error: 2.62 -> 2.49 | Improved |
| add six more unlabeled images from the test camera (m=1 to m=7) (ablation) | C5 · Cube+ ablation (7 encoders, trained with augmentation) | mean angular error: 2.60 -> 1.87 | Improved |
| use more additional images beyond m=7 (m=13) (ablation) | C5 · Cube+ ablation (7 encoders, trained with augmentation) | mean angular error: 1.87 -> 1.95 | No clear effect |
| raise histogram bin size from 16 to 64 (ablation) | C5 · Cube+ ablation (7 encoders, trained with augmentation) | mean angular error: 2.28 -> 1.87 | Improved |
| raise histogram bin size from 64 to 128 (ablation) | C5 · Cube+ ablation (7 encoders, trained with augmentation) | mean angular error: 1.87 -> 2.03 | Hurt |
| choose additional images as vivid instead of random (ablation) | C5 · Cube+ Challenge | mean angular error: 2.24 -> 2.19 | Improved |
Closest papers by meaning
- CCMNet: Leveraging Calibrated Color Correction Matrices for Cross-Camera Color Constancy (ICCV 2025 · similarity 0.89)
- White-Balance First, Adjust Later: Cross-Camera Color Constancy via Vision-Language Evaluation (CVPR 2026 · similarity 0.89)
- FC4: Fully Convolutional Color Constancy With Confidence-Weighted Pooling (CVPR 2017 · similarity 0.88)
- Transfer Learning for Color Constancy via Statistic Perspective (AAAI 2022 · similarity 0.88)
- GCC: Generative Color Constancy via Diffusing a Color Checker (CVPR 2025 · similarity 0.86)
- Attentive Illumination Decomposition Model for Multi-Illuminant White Balancing (CVPR 2024 · similarity 0.83)
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.
