A deep unfolding framework is proposed that addresses this fundamental algorithmic challenge through three contributions: a learned gradient refinement module with wavelength-adaptive step sizes generated from a per-wavelength PSF embedding, a PSF-conditioned penalty estimator, and a Monte Carlo PSF training strategy that improves robustness to manufacturing-induced PSF variations.
Abstract
Deep unfolding methods achieve state-of-the-art performance in coded aperture snapshot spectral imaging (CASSI) reconstruction but typically rely on closed-form data-fidelity updates that assume an ideal optical system. In practical CASSI systems, the field-dependent and wavelength-dependent point spread function (PSF) introduces convolutional coupling that breaks the diagonal structure of the normal equation, rendering closed-form updates inapplicable. We propose a deep unfolding framework that addresses this fundamental algorithmic challenge through three contributions: (1)~a $K$-step conjugate gradient (CG) unrolling that explicitly solves the non-diagonal normal equation under PSF-inclusive forward operators; (2)~a learned gradient refinement module with wavelength-adaptive step sizes generated from a per-wavelength PSF embedding; and (3)~a PSF-conditioned penalty estimator that adapts the ADMM regularization strength to the optical degradation severity. A Monte Carlo PSF training strategy further improves robustness to manufacturing-induced PSF variations. Our method achieves 30.53~dB on the KAIST dataset, +2.73~dB over the DPU baseline (1.27M) using a comparable number of parameters (1.42M), and +1.70~dB over a larger baseline (DPU-B+, 2.12M) using 33\% fewer parameters.
Experiments on AAPM and DeepLesion under multiple sparse-view and noise settings show that CG-GLORE achieves strong quantitative performance, stable convergence, lower noise power, and improved visual fidelity compared with representative reconstruction methods.
Tran Xuan Hieu Le, D. C. Bui, V. Le et al.· 0 citations
A model-driven bilevel optimization framework that couples SENSE-based image reconstruction with SPIRiT-based k-space calibration through shared CSMs, and introduces a deep-prior-guided regularization strategy that preserves the structure of classical linear regularizers while adaptively learning spatially varying regularization weights from denoised intermediate reconstructions.
Weipeng Chen, Yan-Ran Li, Raymond H. Chan et al.· Journal of Mathematical Imag...· 0 citations
Point spread function (PSF) engineering is a promising approach for passive, snapshot 3D imaging with a single detector. A widely used technique is the double-helix PSF (DH-PSF), which employs a specialized phase mask at the pupil plane to modulate incident light, generating rotationally varying PSFs with defocus. By leveraging a precalibrated depth-dependent PSF model, the depth information of the target surface can be recovered from a snapshot measurement. However, existing reconstruction algorithms often lack efficiency and accuracy, primarily due to the block-wise processing of conventional methods or the failure to incorporate physical priors in end-to-end neural networks. To address these limitations, we propose a physics-guided deep unfolding network (PG-DUN) for snapshot 3D imaging with DH-PSFs. By explicitly embedding the imaging model into the deep neural network, our DUN can naturally reconstruct the 2D image and depth map simultaneously, contributing to more accurate and efficient reconstruction than previous approaches. The feasibility and effectiveness of the proposed method are validated through extensive experiments on simulated and real-world data. The proposed method can serve as a prototype for a deep learning-based reconstruction model in similar deconvolution tasks. Its key innovation—an accelerated deconvolutional gradient descent design—functions as a plug-and-play component that enhances the reconstruction accuracy of any deep neural network with negligible added computational cost.
This paper proposes a novel learned spherical alternating direction method of multipliers (LSADMMs) for effective Rician noise removal under spherical constraints, and incorporates a noise-level estimation prefix that provides adaptive guidance across noise levels.
Jun Shi, Zhifang Liu, Chunlin Wu et al.· Inverse Problems· 0 citations
This work proposes UMPIRE-Net (Unrolled Magnitude-Phase In REgularization Network), a PD-DL method that introduces separate learned regularizers for magnitude and phase components, together with a novel data-fidelity formulation that enforces measurements consistency.
Mahdi Saberi, Toygan Kilic, Mehmet Akçakaya· 2 citations
End-to-end metalens-based snapshot spectral imaging jointly optimizes the optical encoder and computational decoder, offering a promising approach for compact and efficient hyperspectral acquisition. However, existing reconstruction methods often provide limited consistency with the known forward imaging model and insufficient modeling of the global frequency-domain dependencies introduced by diffraction-coded convolution, limiting reconstruction accuracy and crossdataset generalization. To address these issues, we propose a physical-consistency-guided deep unfolding method for metalens-based snapshot spectral imaging. The proposed framework embeds the metalens point spread functions and camera spectral response functions into the unfolding reconstruction process, and introduces a physical consistency loss composed of three forward-model consistency terms: PSF-encoding consistency, SRF-projection consistency, and measurement consistency. In addition, we design a spatial–Fourier hybrid prior module as the data-driven prior module in the deep unfolding network. The spatial branch enhances local textures and edge structures, while the Fourier branch captures long-range dependencies and global frequency-domain characteristics introduced by diffraction-coded convolution; the resulting spatial and Fourier features are adaptively fused through an attention mechanism to enhance spectral–spatial feature representation. Trained on the ICVL dataset and evaluated on the NTIRE 2022 HSI dataset, the proposed framework achieves the best reconstruction performance among representative purely deep-learning-based methods for metalens snapshot spectral imaging. This metalens-based snapshot spectral imaging framework advances the development of miniaturized and high-speed spectral imaging systems.
Unknown authors· AI Photonics Technology Symp...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
