On the Global Photometric Alignment for Low-Level Vision
2026-04-09 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors explain that training image restoration models is tricky because differences in brightness or color between paired images confuse the learning process. They show mathematically that this mismatch wastes most of the training effort on adjusting color rather than fixing the image content. To fix this, they introduce a new loss called Photometric Alignment Loss (PAL) that adjusts for these color differences in a smart, efficient way. Testing PAL on many tasks and datasets, the authors find it helps models restore images more accurately and generalize better.
low-level visionpixel-wise lossphotometric inconsistencyimage restorationleast-squares decompositiongradient energyaffine color alignmentloss functiongeneralizationcovariance matrix
Authors
Mingjia Li, Tianle Du, Hainuo Wang, Qiming Hu, Xiaojie Guo
Abstract
Supervised low-level vision models rely on pixel-wise losses against paired references, yet paired training sets exhibit per-pair photometric inconsistency, say, different image pairs demand different global brightness, color, or white-balance mappings. This inconsistency enters through task-intrinsic photometric transfer (e.g., low-light enhancement) or unintended acquisition shifts (e.g., de-raining), and in either case causes an optimization pathology. Standard reconstruction losses allocate disproportionate gradient budget to conflicting per-pair photometric targets, crowding out content restoration. In this paper, we investigate this issue and prove that, under least-squares decomposition, the photometric and structural components of the prediction-target residual are orthogonal, and that the spatially dense photometric component dominates the gradient energy. Motivated by this analysis, we propose Photometric Alignment Loss (PAL). This flexible supervision objective discounts nuisance photometric discrepancy via closed-form affine color alignment while preserving restoration-relevant supervision, requiring only covariance statistics and tiny matrix inversion with negligible overhead. Across 6 tasks, 16 datasets, and 16 architectures, PAL consistently improves metrics and generalization. The implementation is in the appendix.