VGenNet: Variable Generative Prior Enhanced Single Pixel Imaging

Abstract

Single-pixel imaging (SPI) is an emerging imaging methodology that converts a two-or even three-dimensional image acquisition problem into a one-dimensional (1D) temporal-signal detection problem. Thus, it is crucially important to develop efficient SPI techniques for image reconstruction from the 1D measurements, in particular, an undersampled one. Recently, various studies have demonstrated the superiority of deep learning for SPI. However, due to the generalization issue, conventional datadriven deep learning is a task-specific approach. One needs to retrain the neural network for different SPI imaging problems and different types of objects. Here, we propose a variable generative network enhanced SPI algorithm (VGenNet) by incorporating a model-driven fine-tuning process into a generative model that may have been trained for other tasks. VGenNet simultaneously updates the input vector and the weights in a generator to generate feasible solutions that reproduce the raw measurements. We demonstrate the proposed technique with indoor SPI and outdoor 3D single-pixel LiDAR experiments. Our results show that high-quality images can be reconstructed at low sampling ratios under different system configurations, demonstrating the good performance and flexibility of VGenNet. Overall, the proposed VGenNet is a general framework to take advantage of both the data and physics priors, allowing the direct use of a pretrained generative model to solve various inverse imaging problems.

The use of deep learning for computational optical imaging: from data driven to physics driven

“…Instead, we can incorporate a pre-trained generative adversarial network (GAN) into PhysenNet to solve a given problem that is not what that GAN was trained for. 23 For example, we have demonstrated the incorporation of BigGAN 24 into PhysenNet for image reconstruction in GI. 23 BigGAN 24 has rich statistical properties and is widely employed in the community of computer vision.…”

“…23 For example, we have demonstrated the incorporation of BigGAN 24 into PhysenNet for image reconstruction in GI. 23 BigGAN 24 has rich statistical properties and is widely employed in the community of computer vision.…”

Spectral Manipulations of Random Fiber Lasers: Principles, Characteristics, and Applications

“…Particularly, significant advancements in the spectral properties of RFLs have emerged in the aspects of numerical modeling and physical mechanism investigation, lasing wavelength extension, spectrum manipulation, and the applications of spectral-tailored RFLs, etc. [14,[26][27][28][29][30][31][32][33][34][35] Notably, RFLs have gained a solid reputation as being wavelength-flexible lasers capable of generating random lasing within a broad wavelength band benefiting from different gain mechanisms, broadband Rayleigh backscattering feedback, and simple structure design. In the beginning, RFLs were designed to obtain gain from stimulated Raman scattering (SRS) [7,36,37] and stimulated Brillouin scattering (SBS) [38][39][40] in passive fibers.…”

Compressive Reconstruction Based on Sparse Autoencoder Network Prior for Single-Pixel Imaging

“…In order to stabilize the reconstruction, additional constraints and prior knowledge are required. In the early stage, the development of single-photon image reconstruction methods mainly came from the regularization constraints based on compressed sensing theory [5] and the introduction of explicit priors [6], such as one-norm prior [7], Bayesian prior [8], etc. The use of sparsity and low-rank prior [9] information can quickly reconstruct images at low sampling rates, and good reconstruction effects can be obtained.…”