Lida Li
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Lida Li has a total of 40 co-authors. Here are the top co-authors by number of shared publications:
- Hailiang Chen
Yanshan University
6 shared publications - Shuguang Li
Yanshan University
5 shared publications - Lei Zhang
Northwestern Polytechnical University, Xi'an, China
4 shared publications - Mingqi Gu
State Key Laboratory of Metastable Materials Science and Technology, Key Laboratory for Microstructural Material Physics of Hebei Province, School of Science, Yanshan University, Qinhuangdao, China
3 shared publications - Ruyue Shi
Yanshan University
3 shared publications - Sa Zhang
State Key Laboratory of Metastable Materials Science and Technology and the Key Laboratory for Microstructural Material Physics of Hebei Province, School of Science, Yanshan University, Qinhuangdao, China
3 shared publications
Latest Publications
- 15 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 48 (3) , 2170-2185
High-Resolution Photo Enhancement in Real-Time: A Laplacian Pyramid Network
Photo enhancement plays a crucial role in augmenting the visual aesthetics of a photograph. In recent years, photo enhancement methods have either focused on enhancement performance, producing powerful models that cannot be deployed on edge devices, or prioritized computational efficiency, resulting in inadequate performance for real-world applications. To this end, this paper introduces a pyramid network called LLF-LUT++, which integrates global and local operators through closed-form Laplacian pyramid decomposition and reconstruction. This approach enables fast processing of high-resolution images while also achieving excellent performance. Specifically, we utilize an image-adaptive 3D LUT that capitalizes on the global tonal characteristics of downsampled images, while incorporating two distinct weight fusion strategies to achieve coarse global image enhancement. To implement this strategy, we designed a spatial-frequency transformer weight predictor that effectively extracts the desired distinct weights by leveraging frequency features. Additionally, we apply local Laplacian filters to adaptively refine edge details in high-frequency components. After meticulously redesigning the network structure and transformer model, LLF-LUT++ not only achieves a 2.64 dB improvement in PSNR on the HDR+ dataset, but also further reduces runtime, with 4 K resolution images processed in just 13 ms on a single GPU. Extensive experimental results on two benchmark datasets further show that the proposed approach performs favorably compared to state-of-the-art methods.
- 15 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vol. 48 (3) , 2170-2185
High-Resolution Photo Enhancement in Real-Time: A Laplacian Pyramid Network
Photo enhancement plays a crucial role in augmenting the visual aesthetics of a photograph. In recent years, photo enhancement methods have either focused on enhancement performance, producing powerful models that cannot be deployed on edge devices, or prioritized computational efficiency, resulting in inadequate performance for real-world applications. To this end, this paper introduces a pyramid network called LLF-LUT++, which integrates global and local operators through closed-form Laplacian pyramid decomposition and reconstruction. This approach enables fast processing of high-resolution images while also achieving excellent performance. Specifically, we utilize an image-adaptive 3D LUT that capitalizes on the global tonal characteristics of downsampled images, while incorporating two distinct weight fusion strategies to achieve coarse global image enhancement. To implement this strategy, we designed a spatial-frequency transformer weight predictor that effectively extracts the desired distinct weights by leveraging frequency features. Additionally, we apply local Laplacian filters to adaptively refine edge details in high-frequency components. After meticulously redesigning the network structure and transformer model, LLF-LUT++ not only achieves a 2.64 dB improvement in PSNR on the HDR+ dataset, but also further reduces runtime, with 4 K resolution images processed in just 13 ms on a single GPU. Extensive experimental results on two benchmark datasets further show that the proposed approach performs favorably compared to state-of-the-art methods.
- 5 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-7
Label-Free In Situ Detection of DNA Hybridization Based on Long-Range Surface Plasmon Resonance in Microstructured Optical Fiber
Deoxyribonucleic acid (DNA), which is considered to be the gene code relating to some serious diseases such as breast cancer, lymph cancer, leukemia, intestinal cancer and lung cancer, is difficult to detect due to its low concentration in the early stage of diseases. In this paper, we propose a composite film of Au/MgF2 coated pomelo-type microstructured optical fiber (MOF) to detect the low concentration DNA. The dielectric film of MgF2 helped to enhance the penetration depth of surface plasmon polaritons and to achieve the long-range surface plasmon resonance (LRSPR) which was utilized to detect the micron-size molecule of DNA. The composite film-coated MOF immersed in mercaptoethylamine (MEA) solution to aminate and then functionalized with COOH-modified probe DNA (pDNA) to fabricate the sensing probe. Experimental results showed that the fabricated sensing probe was sensitive to the target DNA (tDNA) with an ultra-low concentration of 2 μ M, but insensitive to the non-target DNA (nonDNA). The proposed sensor, showing the merits of label-free, in-situ, real-time, specificity, stability, and high sensitivity, could be a competitive candidate in DNA detection.
- 5 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-7
Label-Free In Situ Detection of DNA Hybridization Based on Long-Range Surface Plasmon Resonance in Microstructured Optical Fiber
Deoxyribonucleic acid (DNA), which is considered to be the gene code relating to some serious diseases such as breast cancer, lymph cancer, leukemia, intestinal cancer and lung cancer, is difficult to detect due to its low concentration in the early stage of diseases. In this paper, we propose a composite film of Au/MgF2 coated pomelo-type microstructured optical fiber (MOF) to detect the low concentration DNA. The dielectric film of MgF2 helped to enhance the penetration depth of surface plasmon polaritons and to achieve the long-range surface plasmon resonance (LRSPR) which was utilized to detect the micron-size molecule of DNA. The composite film-coated MOF immersed in mercaptoethylamine (MEA) solution to aminate and then functionalized with COOH-modified probe DNA (pDNA) to fabricate the sensing probe. Experimental results showed that the fabricated sensing probe was sensitive to the target DNA (tDNA) with an ultra-low concentration of 2 μ M, but insensitive to the non-target DNA (nonDNA). The proposed sensor, showing the merits of label-free, in-situ, real-time, specificity, stability, and high sensitivity, could be a competitive candidate in DNA detection.
- 17 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-10
Highly Sensitive and Ultra-Wide Detection Range Refractive Index Sensor Around a Singular Point in Mach–Zehnder Interference
Refractive index (RI) optical fiber sensors are often used in biomedical and petrochemical fields. However, existing optical fiber sensors can only detect liquids with low RI, and cannot meet the detection needs of high RI liquids such as oil, hydrocarbons and benzene. Due to the limitation of core RI, the detection range of fiber RI sensor based on surface plasmon resonance (SPR) effect, fiber Bragg grating (FBG), Fabry–Pérot interferometer (FPI) and optical waveguide principle is limited to 1.45. It is essential to develop a highly sensitive sensor that can detect liquids with ultra-wide RI. In this paper, a RI fiber sensor based on Mach-Zehnder interferometer (MZI) with lateral offset composite structure is proposed and experimentally verified. MZI was fabricated through precisely cutting and lateral offset splicing between common single-mode fiber (SMF) and multi-mode fiber (MMF). High sensitivity and ultra-wide RI range of 1.33-1.50 near MZI singularities are realized. To our best knowledge, this is the first report of continuous detection of an ultra-wide RI range from 1.33 to 1.50 in experiments using fiber optic sensors. In this range, the transmission spectrum is first blue shifted and then red shifted. There is a highly amplified sensitivity peak at the singular point of the curve. The sensor has the advantages of wide RI detection range, good recovery characteristics, high stability and low production cost, making it a strong competitor in the field of wide range and high sensitivity RI sensing.
- 17 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 74 (00189456) , 1-10
Highly Sensitive and Ultra-Wide Detection Range Refractive Index Sensor Around a Singular Point in Mach–Zehnder Interference
Refractive index (RI) optical fiber sensors are often used in biomedical and petrochemical fields. However, existing optical fiber sensors can only detect liquids with low RI, and cannot meet the detection needs of high RI liquids such as oil, hydrocarbons and benzene. Due to the limitation of core RI, the detection range of fiber RI sensor based on surface plasmon resonance (SPR) effect, fiber Bragg grating (FBG), Fabry–Pérot interferometer (FPI) and optical waveguide principle is limited to 1.45. It is essential to develop a highly sensitive sensor that can detect liquids with ultra-wide RI. In this paper, a RI fiber sensor based on Mach-Zehnder interferometer (MZI) with lateral offset composite structure is proposed and experimentally verified. MZI was fabricated through precisely cutting and lateral offset splicing between common single-mode fiber (SMF) and multi-mode fiber (MMF). High sensitivity and ultra-wide RI range of 1.33-1.50 near MZI singularities are realized. To our best knowledge, this is the first report of continuous detection of an ultra-wide RI range from 1.33 to 1.50 in experiments using fiber optic sensors. In this range, the transmission spectrum is first blue shifted and then red shifted. There is a highly amplified sensitivity peak at the singular point of the curve. The sensor has the advantages of wide RI detection range, good recovery characteristics, high stability and low production cost, making it a strong competitor in the field of wide range and high sensitivity RI sensing.
- 3 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (8) , 12891-12898
Sensitivity-Enhanced Ethanol Vapor Sensor Based on Vernier Effect in Sagnac Interferometers With a Polydimethylsiloxane Film-Coated Polarization-Maintaining Fiber
Fiber-optic interferometer has been employed in volatile organic compound (VOC) sensing owing to its electromagnetic immunity, reliability, and precise monitoring. However, the measurement sensitivity still needs to be further enhanced to meet higher requirement in trace VOC detection. In this paper, fiber-optic Sagnac interferometers (SIs) based on vernier effect (VE) are proposed and experimentally studied to achieve the highly sensitive measurement of ethanol vapor. Two polarization-maintaining fibers (PMFs) with slightly different lengths are used to form the sensing arm and reference arm in the two SIs and thus VE is excited. Polydimethylsiloxane (PDMS) film which is coated on the outside of sensing arm acts as absorbing material for ethanol vapor. The above method resulting in significant enhancement of the measurement sensitivity. Experimental results show that the measurement sensitivities based on cascaded and paralleled SIs are 11.27 pm/ppm and 11.34 pm/ppm respectively, which is enhanced by about 13 times contrasted to the measurement sensitivity based on single SI. Meanwhile, limit of detection (LOD) for the cascaded and paralleled SIs are as low as 2.049 ppm and 2.156 ppm respectively. The proposed sensor has the advantages of high sensitivity, low LOD, good repeatability, strong robustness, and low production cost, showing potential applications in ethanol vapor detection.
- 3 March 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (8) , 12891-12898
Sensitivity-Enhanced Ethanol Vapor Sensor Based on Vernier Effect in Sagnac Interferometers With a Polydimethylsiloxane Film-Coated Polarization-Maintaining Fiber
Fiber-optic interferometer has been employed in volatile organic compound (VOC) sensing owing to its electromagnetic immunity, reliability, and precise monitoring. However, the measurement sensitivity still needs to be further enhanced to meet higher requirement in trace VOC detection. In this paper, fiber-optic Sagnac interferometers (SIs) based on vernier effect (VE) are proposed and experimentally studied to achieve the highly sensitive measurement of ethanol vapor. Two polarization-maintaining fibers (PMFs) with slightly different lengths are used to form the sensing arm and reference arm in the two SIs and thus VE is excited. Polydimethylsiloxane (PDMS) film which is coated on the outside of sensing arm acts as absorbing material for ethanol vapor. The above method resulting in significant enhancement of the measurement sensitivity. Experimental results show that the measurement sensitivities based on cascaded and paralleled SIs are 11.27 pm/ppm and 11.34 pm/ppm respectively, which is enhanced by about 13 times contrasted to the measurement sensitivity based on single SI. Meanwhile, limit of detection (LOD) for the cascaded and paralleled SIs are as low as 2.049 ppm and 2.156 ppm respectively. The proposed sensor has the advantages of high sensitivity, low LOD, good repeatability, strong robustness, and low production cost, showing potential applications in ethanol vapor detection.
- 21 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 33 (10577149) , 6114-6128
Toward Blind Flare Removal Using Knowledge-Driven Flare-Level Estimator
Lens flare is a common phenomenon when strong light rays arrive at the camera sensor and a clean scene is consequently mixed up with various opaque and semi-transparent artifacts. Existing deep learning methods are always constrained with limited real image pairs for training. Though recent synthesis-based approaches are found effective, synthesized pairs still deviate from the real ones as the mixing mechanism of flare artifacts and scenes in the wild always depends on a line of undetermined factors, such as lens structure, scratches, etc . In this paper, we present a new perspective from the blind nature of the flare removal task in a knowledge-driven manner. Specifically, we present a simple yet effective flare-level estimator to predict the corruption level of a flare-corrupted image. The estimated flare-level can be interpreted as additive information of the gap between corrupted images and their flare-free correspondences to facilitate a network at both training and testing stages adaptively. Besides, we utilize a flare-level modulator to better integrate the estimations into networks. We also devise a flare-aware block for more accurate flare recognition and reconstruction. Additionally, we collect a new real-world flare dataset for benchmarking, namely WiderFlare. Extensive experiments on three benchmark datasets demonstrate that our method outperforms state-of-the-art methods quantitatively and qualitatively.
- 21 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Image Processing
- Vol. 33 (10577149) , 6114-6128
Toward Blind Flare Removal Using Knowledge-Driven Flare-Level Estimator
Lens flare is a common phenomenon when strong light rays arrive at the camera sensor and a clean scene is consequently mixed up with various opaque and semi-transparent artifacts. Existing deep learning methods are always constrained with limited real image pairs for training. Though recent synthesis-based approaches are found effective, synthesized pairs still deviate from the real ones as the mixing mechanism of flare artifacts and scenes in the wild always depends on a line of undetermined factors, such as lens structure, scratches, etc . In this paper, we present a new perspective from the blind nature of the flare removal task in a knowledge-driven manner. Specifically, we present a simple yet effective flare-level estimator to predict the corruption level of a flare-corrupted image. The estimated flare-level can be interpreted as additive information of the gap between corrupted images and their flare-free correspondences to facilitate a network at both training and testing stages adaptively. Besides, we utilize a flare-level modulator to better integrate the estimations into networks. We also devise a flare-aware block for more accurate flare recognition and reconstruction. Additionally, we collect a new real-world flare dataset for benchmarking, namely WiderFlare. Extensive experiments on three benchmark datasets demonstrate that our method outperforms state-of-the-art methods quantitatively and qualitatively.
- 16 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Lightwave Technology
- Vol. 43 (3) , 1429-1436
A High Resolution and Large Measurement Range Strain Sensor Based on Sagnac Loop and Fiber Bragg Grating in Dual-Wavelength Erbium-Doped Fiber Laser
The high-resolution strain detection holds significant research value in terms of providing early warning for cracks in aircraft, wind energy blades, and combustible gas containers. We propose and experimentally demonstrate a strain sensor based on Sagnac loop and Fiber Bragg grating (FBG) in dual-wavelength erbium-doped fiber laser (EDFL), showing the properties of high resolution and wide detection range, the polarization maintaining fiber (PMF) in the Sagnac loop serves as both an interference filter and a sensing probe. Nonetheless, the strain measurement range is limited by the Sagnac loop's short free spectral range, we add an FBG in the Sagnac loop, which provides a calibration element to improve the range. We successfully improve the spectral quality factor to 170.56 and reach a high resolution of 0.067 nm in a wide measurement range of 0-6434.81 µϵ. The results of our experiments show that the method is ideally suited for precise strain measurement, with major potential applications in environmental monitoring, health monitoring, biomedicine, and aerospace.
- 16 October 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in Journal of Lightwave Technology
- Vol. 43 (3) , 1429-1436
A High Resolution and Large Measurement Range Strain Sensor Based on Sagnac Loop and Fiber Bragg Grating in Dual-Wavelength Erbium-Doped Fiber Laser
The high-resolution strain detection holds significant research value in terms of providing early warning for cracks in aircraft, wind energy blades, and combustible gas containers. We propose and experimentally demonstrate a strain sensor based on Sagnac loop and Fiber Bragg grating (FBG) in dual-wavelength erbium-doped fiber laser (EDFL), showing the properties of high resolution and wide detection range, the polarization maintaining fiber (PMF) in the Sagnac loop serves as both an interference filter and a sensing probe. Nonetheless, the strain measurement range is limited by the Sagnac loop's short free spectral range, we add an FBG in the Sagnac loop, which provides a calibration element to improve the range. We successfully improve the spectral quality factor to 170.56 and reach a high resolution of 0.067 nm in a wide measurement range of 0-6434.81 µϵ. The results of our experiments show that the method is ideally suited for precise strain measurement, with major potential applications in environmental monitoring, health monitoring, biomedicine, and aerospace.
- 31 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-10
A Wide Range Curvature Sensor Based on Fiber Ring Laser With a Long-Period Fiber Grating
Long-Period Fiber Grating (LPFG) is a promising curvature sensor due to its high bending sensitivity. In this paper, a Fiber Ring Laser (FRL)-based curvature sensor with an LPFG was proposed and experimentally verified. LPFG acted as both a filter and a curvature sensing element. The maximum measurement sensitivity of curvature reached 7.99 dB/m -1 in a wide range of 4.89-13.33 m -1 which was wider than most of previously reported curvature sensors. Bending and filtering characteristics of LPFG were verified through changing its position in FRL. The output wavelength of the FRL could be tuned through the bending of LPFG. The proposed curvature sensor, showing the advantages of wide measurement range, high resolution, and high sensitivity, has potential applications in monitoring the deformation in surgical instruments, robot arm, and engineering objects.
- 31 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-10
A Wide Range Curvature Sensor Based on Fiber Ring Laser With a Long-Period Fiber Grating
Long-Period Fiber Grating (LPFG) is a promising curvature sensor due to its high bending sensitivity. In this paper, a Fiber Ring Laser (FRL)-based curvature sensor with an LPFG was proposed and experimentally verified. LPFG acted as both a filter and a curvature sensing element. The maximum measurement sensitivity of curvature reached 7.99 dB/m -1 in a wide range of 4.89-13.33 m -1 which was wider than most of previously reported curvature sensors. Bending and filtering characteristics of LPFG were verified through changing its position in FRL. The output wavelength of the FRL could be tuned through the bending of LPFG. The proposed curvature sensor, showing the advantages of wide measurement range, high resolution, and high sensitivity, has potential applications in monitoring the deformation in surgical instruments, robot arm, and engineering objects.
- 15 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-10
Highly Sensitive Strain and Temperature Sensors Based on Vernier Effect in Cascaded Mach–Zehnder Interferometer and Sagnac Interferometer
In this article, vernier effect (VE), which is an effective method to enhance the measurement sensitivity of strain and temperature, was experimentally studied through cascading a Mach-Zehnder interferometer (MZI) with a Sagnac interferometer (SI). The MZI consisted of a tapered single-mode fiber (SMF), while the SI was formed using a panda-shaped polarization-maintained fiber (PMF). Free spectral ranges (FSRs) of the two interferometers were slightly different from each other, and as a result, the VE was successfully excited. The experimental results demonstrated that the measurement sensitivity of strain was improved from 26.7 to −110.92 pm/$\mu \varepsilon $ at the temperature of$24~^{\circ }$ C, while the measurement sensitivity of temperature was improved from −2978 to 7934 pm/°C at the strain of$0~\varepsilon $ . The sensitivity was amplified about four times. The cross-sensitivity of temperature and strain measurements is$71.53~\mu \varepsilon $ /°C. The proposed sensor exhibits in advantages of simple in construction, high sensitivity, and multiparameter measurements making it a superior contender for strain and temperature monitoring in engineering.- 15 July 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Instrumentation and Measurement
- Vol. 73 (00189456) , 1-10
Highly Sensitive Strain and Temperature Sensors Based on Vernier Effect in Cascaded Mach–Zehnder Interferometer and Sagnac Interferometer
In this article, vernier effect (VE), which is an effective method to enhance the measurement sensitivity of strain and temperature, was experimentally studied through cascading a Mach-Zehnder interferometer (MZI) with a Sagnac interferometer (SI). The MZI consisted of a tapered single-mode fiber (SMF), while the SI was formed using a panda-shaped polarization-maintained fiber (PMF). Free spectral ranges (FSRs) of the two interferometers were slightly different from each other, and as a result, the VE was successfully excited. The experimental results demonstrated that the measurement sensitivity of strain was improved from 26.7 to −110.92 pm/$\mu \varepsilon $ at the temperature of$24~^{\circ }$ C, while the measurement sensitivity of temperature was improved from −2978 to 7934 pm/°C at the strain of$0~\varepsilon $ . The sensitivity was amplified about four times. The cross-sensitivity of temperature and strain measurements is$71.53~\mu \varepsilon $ /°C. The proposed sensor exhibits in advantages of simple in construction, high sensitivity, and multiparameter measurements making it a superior contender for strain and temperature monitoring in engineering.- 1 February 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 33 (8) , 3400-3414
Detachable Second-Order Pooling: Toward High-Performance First-Order Networks
Second-order pooling has proved to be more effective than its first-order counterpart in visual classification tasks. However, second-order pooling suffers from the high demand for a computational resource, limiting its use in practical applications. In this work, we present a novel architecture, namely a detachable second-order pooling network, to leverage the advantage of second-order pooling by first-order networks while keeping the model complexity unchanged during inference. Specifically, we introduce second-order pooling at the end of a few auxiliary branches and plug them into different stages of a convolutional neural network. During the training stage, the auxiliary second-order pooling networks assist the backbone first-order network to learn more discriminative feature representations. When training is completed, all auxiliary branches can be removed, and only the backbone first-order network is used for inference. Experiments conducted on CIFAR-10, CIFAR-100, and ImageNet data sets clearly demonstrated the leading performance of our network, which achieves even higher accuracy than second-order networks but keeps the low inference complexity of first-order networks.
- 1 February 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Neural Networks and Learning Systems
- Vol. 33 (8) , 3400-3414
Detachable Second-Order Pooling: Toward High-Performance First-Order Networks
Second-order pooling has proved to be more effective than its first-order counterpart in visual classification tasks. However, second-order pooling suffers from the high demand for a computational resource, limiting its use in practical applications. In this work, we present a novel architecture, namely a detachable second-order pooling network, to leverage the advantage of second-order pooling by first-order networks while keeping the model complexity unchanged during inference. Specifically, we introduce second-order pooling at the end of a few auxiliary branches and plug them into different stages of a convolutional neural network. During the training stage, the auxiliary second-order pooling networks assist the backbone first-order network to learn more discriminative feature representations. When training is completed, all auxiliary branches can be removed, and only the backbone first-order network is used for inference. Experiments conducted on CIFAR-10, CIFAR-100, and ImageNet data sets clearly demonstrated the leading performance of our network, which achieves even higher accuracy than second-order networks but keeps the low inference complexity of first-order networks.
- 7 November 2020
- Published by Springer Nature
- p. 562-579
LST-Net: Learning a Convolutional Neural Network with a Learnable Sparse Transform
- 7 November 2020
- Published by Springer Nature
- p. 562-579
LST-Net: Learning a Convolutional Neural Network with a Learnable Sparse Transform