Hierarchical Regression Network for Spectral Reconstruction from RGB Images

Zhao, Yuzhi; Po, Lai-Man; Yan, Qing; Liu, Wei; Lin, Ting-Yu

doi:10.1109/cvprw50498.2020.00219

Cited by 108 publications

(87 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The batch size was set as 20, the optimization parameters

and

were 0.9 and 0.99, respectively,

was 10 −8 , the DOFCBAM reduction was 16, the ONB reduction was 8, the learning rate was initialized as 0.0001, and the attenuation strategy of the polynomial function as a power function was equal to 1.5. In the process of comparing the proposed method with the YAN [ 43 ], HRN [ 43 ] and AWAN [ 45 ] algorithms, all the experiments were carried out on the same hardware, programming environment and dataset.…”

Section: Experimental Results and Analysismentioning

confidence: 99%

“…Our method, YAN, AWAN, and HRN [ 43 ] were compared for another NTIRE 2020 hyperspectral validation set for RGB image reconstruction using the same hardware equipment and a batch size set to 20. The visual effect is shown in Figure 9 and the results of the comparison of performances are shown in Table 2 .…”

Section: Experimental Results and Analysismentioning

confidence: 99%

“…The reason for this better performance is that we employ the ghost network framework, which uses convolution kernels (size 1 × 1). In Table 2 ’s comparison of our method with the YAN [ 42 ], AWAN [ 45 ] and HRN [ 43 ] algorithms, the model size of the model trained by this method is the smallest, which is 32,898 kb. The proposed DGCAMN framework model greatly reduces the amount of computer hardware storage during RGB to hyperspectral reconstruction and is a lightweight framework model under conditions that produce the same spectral reconstruction accuracy.…”

Section: Experimental Results and Analysismentioning

confidence: 99%

“…This is conducive to the extraction of the local features of the images, but fewer parameters are calculated. The minimum calculation parameters for our algorithm is 2,783,247 KB, which is 1/5, 4/25, and1/38 of the number for the YAN [ 42 ], AWAN [ 45 ], and the HRN [ 43 ] algorithms, respectively ( Table 2 ).…”

Section: Experimental Results and Analysismentioning

confidence: 99%

“…In a multiscale deep CNN proposed by YAN [ 42 ] through the symmetric cascade down sampling and up sampling of the intermediate feature map, the local and nonlocal image information can be jointly encoded for spectral representation, and the accuracy of spectral reconstruction can be improved. The network that deepens the convolutional layer is the hierarchical regression network (HRN) [ 43 ]. Therefore, residual dense blocks are used to remove image artifacts and global blocks at each level to expand the visual perception of the image.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Double Ghost Convolution Attention Mechanism Network: A Framework for Hyperspectral Reconstruction of a Single RGB Image

Wang

Jiangwei

2021

Sensors

View full text Add to dashboard Cite

Current research on the reconstruction of hyperspectral images from RGB images using deep learning mainly focuses on learning complex mappings through deeper and wider convolutional neural networks (CNNs). However, the reconstruction accuracy of the hyperspectral image is not high and among other issues the model for generating these images takes up too much storage space. In this study, we propose the double ghost convolution attention mechanism network (DGCAMN) framework for the reconstruction of a single RGB image to improve the accuracy of spectral reconstruction and reduce the storage occupied by the model. The proposed DGCAMN consists of a double ghost residual attention block (DGRAB) module and optimal nonlocal block (ONB). DGRAB module uses GhostNet and PRELU activation functions to reduce the calculation parameters of the data and reduce the storage size of the generative model. At the same time, the proposed double output feature Convolutional Block Attention Module (DOFCBAM) is used to capture the texture details on the feature map to maximize the content of the reconstructed hyperspectral image. In the proposed ONB, the Argmax activation function is used to obtain the region with the most abundant feature information and maximize the most useful feature parameters. This helps to improve the accuracy of spectral reconstruction. These contributions enable the DGCAMN framework to achieve the highest spectral accuracy with minimal storage consumption. The proposed method has been applied to the NTIRE 2020 dataset. Experimental results show that the proposed DGCAMN method outperforms the spectral accuracy reconstructed by advanced deep learning methods and greatly reduces storage consumption.

show abstract

“…The batch size was set as 20, the optimization parameters

and

were 0.9 and 0.99, respectively,

Section: Experimental Results and Analysismentioning

confidence: 99%

Section: Experimental Results and Analysismentioning

confidence: 99%

Section: Experimental Results and Analysismentioning

confidence: 99%

Section: Experimental Results and Analysismentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Double Ghost Convolution Attention Mechanism Network: A Framework for Hyperspectral Reconstruction of a Single RGB Image

Wang

Jiangwei

2021

Sensors

View full text Add to dashboard Cite

show abstract

Snapshot Multispectral Light‐Field Imaging Based on Nematic Liquid‐Crystal Microlens Array via Deep Learning

Li,

Chen

et al. 2023

Adv Eng Mater

View full text Add to dashboard Cite

In this study, we have developed a multispectral light field imaging system using nematic liquid crystal microlens arrays (LC‐MLAs). The system solves the spectral under‐sampling and complex system issues in existing light field and spectral cameras. The LC‐MLA is the core optical element of the snapshot imaging system. No dispersion or customized scanning machinery is required to obtain multispectral light fields. Deep learning technology is utilized to reconstruct spectral information from raw data. Experiments demonstrate that the proposed method efficiently reconstructs multispectral data with 10nm accuracy in the visible wavelength range of [400nm, 700nm]. The system offers high precision, simple equipment, and relatively low cost without additional dispersion and filtering devices.This article is protected by copyright. All rights reserved.

show abstract

Auto-Encoder Guided Attention Based Network for Hyperspectral Recovery from Real RGB Images

Shukla,

Sharma,

Bhugra

et al. 2024

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Hierarchical Regression Network for Spectral Reconstruction from RGB Images

Cited by 108 publications

References 29 publications

Double Ghost Convolution Attention Mechanism Network: A Framework for Hyperspectral Reconstruction of a Single RGB Image

Double Ghost Convolution Attention Mechanism Network: A Framework for Hyperspectral Reconstruction of a Single RGB Image

Snapshot Multispectral Light‐Field Imaging Based on Nematic Liquid‐Crystal Microlens Array via Deep Learning

Auto-Encoder Guided Attention Based Network for Hyperspectral Recovery from Real RGB Images

Contact Info

Product

Resources

About