Low-Rank CP Tensor Decomposition for Lightweight Image Super-Resolution
Open Access
- 5 November 2026
- journal article
- research article
- Published by MDPI AG in Mathematics
- Vol. 14 (21) , 4049
Abstract
Deep convolutional neural networks have substantially advanced single-image super-resolution, or SR, but their high parameter counts and computational redundancy hinder deployment on resource-constrained edge devices. Most existing compression methods apply pruning or low-rank truncation as a post hoc step after training. This “train-then-compress” pipeline is reported in the compression literature to disrupt the spectral coherence of the learned feature manifold and to require costly fine-tuning to restore fidelity. We study an alternative, training-time approach that re-parameterizes the convolutional kernel with CANDECOMP/PARAFAC (CP) decomposition into three low-rank sublayers, optimizing directly on a rank-constrained manifold rather than compressing after convergence. We interpret the forward pass as a heuristic realization, rather than a formal equivalence, of a proximal gradient-descent–ADMM update for a tensor nuclear-norm-regularized problem, which motivates the design and its hyperparameters without invoking expensive singular value decompositions. An analysis of parameter complexity and gradient flow justifies the choice of CP over Tucker and tensor-train decompositions. Experiments on DIV2K, Kodak24, BSD100, and Set12 show that the CP module yields a M-parameter lightweight model at parity with a same-budget reduced residual baseline (at most dB difference on every test set), sacrificing roughly half a decibel relative to a larger residual baseline. Applied to a large-capacity SR backbone, the same module removes of the parameters, from M to M, at a PSNR cost of only – dB. These results indicate that training-time CP decomposition is a simple and parameter-efficient route to compact SR models in the extreme-compression regime.
Keywords
Funding Information
- Science and Technology Department of Xinjiang Uyghur Autonomous Region (2025D01B15)
This publication has 29 references indexed in Scilit:
- RNSR: Retentive networks for lightweight image super-resolutionNeurocomputing, 2025
- Reduced storage direct tensor ring decomposition for convolutional neural networks compressionNeural Networks, 2025
- Efficient frequency feature aggregation transformer for image super-resolutionPattern Recognition, 2025
- QuaDCNN: Quantized compression of deep CNN based on tensor-train decomposition with automatic rank determinationNeurocomputing, 2025
- Diffusion Models, Image Super-Resolution, and Everything: A SurveyIEEE Transactions on Neural Networks and Learning Systems, 2024
- Augmented Lagrangian method for tensor low-rank and sparsity models in multi-dimensional image recoveryAdvances in Computational Mathematics, 2024
- Deep Learning Model Compression With Rank Reduction in Tensor DecompositionIEEE Transactions on Neural Networks and Learning Systems, 2023
- Tensor Decomposition for Model Reduction in Neural Networks: A Review [Feature]IEEE Circuits and Systems Magazine, 2023
- Beyond a Gaussian Denoiser: Residual Learning of Deep CNN for Image DenoisingIEEE Transactions on Image Processing, 2017
- Tensor Decompositions and ApplicationsSIAM Review, 2009