Mingxuan Gu
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Mingxuan Gu has a total of 35 co-authors. Here are the top co-authors by number of shared publications:
Liang Wang
5 shared publications- Andreas Maier
Friedrich-Alexander-Universität Erlangen-Nürnberg, Pattern Recognition Lab, Erlangen, Germany
4 shared publications - Fabian Wagner
Energy, Climate and Environment Program, International Institute for Applied Systems Analysis (IIASA), Laxenburg, Austria
3 shared publications Gang Li
3 shared publications- Mareike Thies
Pattern Recognition Lab, FAU Erlangen-Nürnberg
3 shared publications Pengda Shi
3 shared publications
Latest Publications
- Open Access
- 14 March 2026
- Published by Copernicus GmbH
Bayesian-Optimized XGBoost for High-Precision Lithofacies Identification in Carbonate Reservoirs Using Geophysical Well Logs
Carbonate reservoirs possess substantial hydrocarbon resource potential and have become a major focus of exploration in recent years. However, the complex geological settings, pronounced lithological heterogeneity, and strong vertical variability of carbonate reservoirs pose significant challenges to conventional lithofacies identification methods. Such approaches are often time-consuming, highly subjective, and limited in their ability to accurately discriminate complex lithofacies assemblages, thereby constraining the efficient development of carbonate reservoirs. Geophysical well logging offers advantages such as low acquisition cost, continuous coverage, and high vertical resolution, making it a fundamental dataset for continuous lithofacies characterization. In this study, integrated geological information, including whole-rock analysis, petrographic thin sections, and scanning electron microscopy, is employed to systematically investigate the mineral compositions and pore structure characteristics of different lithofacies, and the corresponding logging response mechanisms are quantitatively investigated. Five conventional logging curves—gamma ray, acoustic, neutron porosity, bulk density, and resistivity—are selected to construct a multidimensional feature parameter set. Considering the differences in numerical ranges among logging curves and the imbalance in lithofacies sample distributions, data normalization and class imbalance correction are performed prior to model training. Subsequently, a high-precision lithofacies identification model based on conventional well logs is developed by integrating Bayesian optimization with the Extreme Gradient Boosting (XGBoost) algorithm. The results indicate that the predicted lithofacies are highly consistent with geological interpretations, core observations, and thin-section identifications, demonstrating that the Bayesian-optimized XGBoost model exhibits robust classification performance and effectively captures the complex nonlinear relationships between lithofacies and logging responses. Compared with traditional machine learning methods, the proposed approach achieves significantly improved identification accuracy, providing robust technical support and a reliable theoretical foundation for carbonate reservoir evaluation and hydrocarbon exploration.
- 14 March 2026
- Published by Copernicus GmbH
Open AccessBayesian-Optimized XGBoost for High-Precision Lithofacies Identification in Carbonate Reservoirs Using Geophysical Well Logs
Carbonate reservoirs possess substantial hydrocarbon resource potential and have become a major focus of exploration in recent years. However, the complex geological settings, pronounced lithological heterogeneity, and strong vertical variability of carbonate reservoirs pose significant challenges to conventional lithofacies identification methods. Such approaches are often time-consuming, highly subjective, and limited in their ability to accurately discriminate complex lithofacies assemblages, thereby constraining the efficient development of carbonate reservoirs. Geophysical well logging offers advantages such as low acquisition cost, continuous coverage, and high vertical resolution, making it a fundamental dataset for continuous lithofacies characterization. In this study, integrated geological information, including whole-rock analysis, petrographic thin sections, and scanning electron microscopy, is employed to systematically investigate the mineral compositions and pore structure characteristics of different lithofacies, and the corresponding logging response mechanisms are quantitatively investigated. Five conventional logging curves—gamma ray, acoustic, neutron porosity, bulk density, and resistivity—are selected to construct a multidimensional feature parameter set. Considering the differences in numerical ranges among logging curves and the imbalance in lithofacies sample distributions, data normalization and class imbalance correction are performed prior to model training. Subsequently, a high-precision lithofacies identification model based on conventional well logs is developed by integrating Bayesian optimization with the Extreme Gradient Boosting (XGBoost) algorithm. The results indicate that the predicted lithofacies are highly consistent with geological interpretations, core observations, and thin-section identifications, demonstrating that the Bayesian-optimized XGBoost model exhibits robust classification performance and effectively captures the complex nonlinear relationships between lithofacies and logging responses. Compared with traditional machine learning methods, the proposed approach achieves significantly improved identification accuracy, providing robust technical support and a reliable theoretical foundation for carbonate reservoir evaluation and hydrocarbon exploration.
- Open Access
- 14 March 2026
- Published by Copernicus GmbH
Fracture Identification in Electrical Image Logs with Limited Samples by Incorporating Outcrop Priors
Micro-resistivity imaging logging is one of the primary techniques for subsurface fracture identification. However, conventional manual interpretation is time-consuming and highly subjective, while existing deep learning-based methods generally require large-scale, well-annotated datasets, resulting in substantial labeling costs and limited applicability in data-scarce scenarios. To address these challenges, this study proposes a fracture identification method for electrical image logs under limited-sample conditions by incorporating prior knowledge derived from outcrop fractures. Leveraging the morphological and statistical similarities between surface outcrop fractures and subsurface electrical image responses, a two-stage training strategy based on a Multi-scale Attention Network (MANet) backbone is developed. In the first stage, optical images of outcrop fractures are preprocessed through grayscale transformation and noise injection to approximate the feature distribution of electrical image logs, enabling the network to learn generalizable edge and texture features. In the second stage, outcrop–electrical image pairs with similar fracture morphology and texture characteristics are generated through similarity matching, and the model is fine-tuned using a composite loss function incorporating Correlation Alignment (CORAL), thereby accelerating domain adaptation to subsurface logging environments. Experimental results from basement reservoirs in the Dongping area of the Qaidam Basin demonstrate that the proposed method significantly improves fracture identification performance under limited-sample conditions. Compared with baseline models, the proposed approach achieves improvements of 13.05% in accuracy and 12.88% in Intersection over Union (IoU), reaching 81.13% and 75.74%, respectively. These results indicate that the proposed method effectively alleviates data scarcity issues in electrical image log interpretation and provides robust technical support for fracture characterization and hydrocarbon resource evaluation.
- 14 March 2026
- Published by Copernicus GmbH
Open AccessFracture Identification in Electrical Image Logs with Limited Samples by Incorporating Outcrop Priors
Micro-resistivity imaging logging is one of the primary techniques for subsurface fracture identification. However, conventional manual interpretation is time-consuming and highly subjective, while existing deep learning-based methods generally require large-scale, well-annotated datasets, resulting in substantial labeling costs and limited applicability in data-scarce scenarios. To address these challenges, this study proposes a fracture identification method for electrical image logs under limited-sample conditions by incorporating prior knowledge derived from outcrop fractures. Leveraging the morphological and statistical similarities between surface outcrop fractures and subsurface electrical image responses, a two-stage training strategy based on a Multi-scale Attention Network (MANet) backbone is developed. In the first stage, optical images of outcrop fractures are preprocessed through grayscale transformation and noise injection to approximate the feature distribution of electrical image logs, enabling the network to learn generalizable edge and texture features. In the second stage, outcrop–electrical image pairs with similar fracture morphology and texture characteristics are generated through similarity matching, and the model is fine-tuned using a composite loss function incorporating Correlation Alignment (CORAL), thereby accelerating domain adaptation to subsurface logging environments. Experimental results from basement reservoirs in the Dongping area of the Qaidam Basin demonstrate that the proposed method significantly improves fracture identification performance under limited-sample conditions. Compared with baseline models, the proposed approach achieves improvements of 13.05% in accuracy and 12.88% in Intersection over Union (IoU), reaching 81.13% and 75.74%, respectively. These results indicate that the proposed method effectively alleviates data scarcity issues in electrical image log interpretation and provides robust technical support for fracture characterization and hydrocarbon resource evaluation.
- Open Access
- 13 March 2026
- Published by Copernicus GmbH
Direct Fractal Characterization of Shale Hydrocarbon Content from NMR T1-T2 Echo Train Data
Accurate estimation of hydrocarbon content is a critical component of shale reservoir evaluation. Although nuclear magnetic resonance (NMR) T1-T2 measurements are highly sensitive to fluid properties, conventional assessments of hydrocarbon content typically rely on empirical interpretation charts or supplementary experiments, which limit their quantitative reliability and practical applicability. In this study, we propose a novel fractal characterization method based on NMR T1-T2 measurements for quantitative evaluation of hydrocarbon content in shale reservoirs. To mitigate the uncertainty introduced by T1-T2 spectral inversion, fractal parameters are directly extracted from the original NMR echo train data, bypassing the inversion process entirely. Numerical simulations demonstrate that the echo-based fractal parameters exhibit significantly enhanced sensitivity and discrimination capability with respect to hydrocarbon content when compared with fractal parameters derived from inverted T1-T2 spectra. Core-scale experiments further validate that the proposed fractal dimension effectively differentiates movable hydrocarbons from pyrolytic hydrocarbons in shale formations. The proposed method provides a robust, efficient, and inversion-independent approach for shale hydrocarbon content evaluation, offering strong potential for both laboratory studies and field-scale NMR applications.
- 13 March 2026
- Published by Copernicus GmbH
Open AccessDirect Fractal Characterization of Shale Hydrocarbon Content from NMR T1-T2 Echo Train Data
Accurate estimation of hydrocarbon content is a critical component of shale reservoir evaluation. Although nuclear magnetic resonance (NMR) T1-T2 measurements are highly sensitive to fluid properties, conventional assessments of hydrocarbon content typically rely on empirical interpretation charts or supplementary experiments, which limit their quantitative reliability and practical applicability. In this study, we propose a novel fractal characterization method based on NMR T1-T2 measurements for quantitative evaluation of hydrocarbon content in shale reservoirs. To mitigate the uncertainty introduced by T1-T2 spectral inversion, fractal parameters are directly extracted from the original NMR echo train data, bypassing the inversion process entirely. Numerical simulations demonstrate that the echo-based fractal parameters exhibit significantly enhanced sensitivity and discrimination capability with respect to hydrocarbon content when compared with fractal parameters derived from inverted T1-T2 spectra. Core-scale experiments further validate that the proposed fractal dimension effectively differentiates movable hydrocarbons from pyrolytic hydrocarbons in shale formations. The proposed method provides a robust, efficient, and inversion-independent approach for shale hydrocarbon content evaluation, offering strong potential for both laboratory studies and field-scale NMR applications.
- Open Access
- 13 March 2026
- Published by Copernicus GmbH
An NMR EMG-Based Method for Pore Structure Characterization and Permeability Prediction in Carbonate Reservoirs
Carbonate reservoirs host substantial hydrocarbon resources; however, their characterization remains challenging due to strong heterogeneity and complex pore systems, which often produce asymmetric and highly multimodal NMR T₂ distributions. These characteristics undermine the applicability and robustness of conventional pore-structure interpretation and permeability models. To overcome these limitations, we propose a novel nuclear magnetic resonance (NMR)-based method that employs an Exponentially Modified Gaussian (EMG) model to quantitatively characterize pore structure and improve permeability estimation. First, the EMG function is used to decompose the measured T₂ distribution into multiple components with clear physical implications, enabling separation and quantification of pore contributions across scales. Second, the EMG-derived characteristic parameters are subsequently incorporated to refine the conventional Schlumberger-Doll Research (SDR) permeability model, thereby accounting for the impact of complex pore geometry on flow capacity. Validation using diverse carbonate samples demonstrates that the EMG model provides accurate and stable fitting of T₂ distributions across varying pore complexity, ranging from unimodal to highly multimodal distributions. Moreover, EMG-informed permeability estimation yields significantly improved accuracy and robustness compared with conventional methods. Overall, the proposed NMR EMG-based method offers a reliable solution for pore-structure characterization and permeability evaluation in complex carbonate reservoirs.
- 13 March 2026
- Published by Copernicus GmbH
Open AccessAn NMR EMG-Based Method for Pore Structure Characterization and Permeability Prediction in Carbonate Reservoirs
Carbonate reservoirs host substantial hydrocarbon resources; however, their characterization remains challenging due to strong heterogeneity and complex pore systems, which often produce asymmetric and highly multimodal NMR T₂ distributions. These characteristics undermine the applicability and robustness of conventional pore-structure interpretation and permeability models. To overcome these limitations, we propose a novel nuclear magnetic resonance (NMR)-based method that employs an Exponentially Modified Gaussian (EMG) model to quantitatively characterize pore structure and improve permeability estimation. First, the EMG function is used to decompose the measured T₂ distribution into multiple components with clear physical implications, enabling separation and quantification of pore contributions across scales. Second, the EMG-derived characteristic parameters are subsequently incorporated to refine the conventional Schlumberger-Doll Research (SDR) permeability model, thereby accounting for the impact of complex pore geometry on flow capacity. Validation using diverse carbonate samples demonstrates that the EMG model provides accurate and stable fitting of T₂ distributions across varying pore complexity, ranging from unimodal to highly multimodal distributions. Moreover, EMG-informed permeability estimation yields significantly improved accuracy and robustness compared with conventional methods. Overall, the proposed NMR EMG-based method offers a reliable solution for pore-structure characterization and permeability evaluation in complex carbonate reservoirs.
- Open Access
- 13 March 2026
- Published by Copernicus GmbH
A Rock-Constrained Multilevel GMM Approach for Wellsite NMR Multifluid Quantitative Evaluation in Shale
Identification of fluid components and reliable saturation evaluation remain critical challenges in shale exploration and development. Conventional core experimental methods are often limited by high costs and long cycle times, while laboratory nuclear magnetic resonance (NMR) techniques fail to perform rapid and continuous measurements. In this study, we propose a rock-constrained multilevel Gaussian mixture model (GMM) approach for the quantitative evaluation of multiple fluids in shale using wellsite NMR data. The proposed workflow begins with normalization and thresholding of the T₁–T₂ spectra. A rock-physics constraint is then incorporated to partition the relaxation domain and distinguish movable fluids from bound fluids. GMM clustering is first applied independently within each relaxation region to extract representative fluid signatures. These characteristic signatures are subsequently integrated into a second-stage GMM analysis, enabling robust identification and quantitative evaluation of individual fluid components. Application to representative shale NMR datasets, coupled with multi-state core experiments, demonstrates that the proposed method effectively identifies multiple fluid components, including heavy components, bound oil, and movable oil. Furthermore, the saturation of various hydrocarbon components can be quantitatively predicted, demonstrating strong agreement with Rock-Eval measurements. The proposed approach provides a practical and reliable solution for rapid downhole fluid identification and saturation evaluation in shale reservoirs.
- 13 March 2026
- Published by Copernicus GmbH
Open AccessA Rock-Constrained Multilevel GMM Approach for Wellsite NMR Multifluid Quantitative Evaluation in Shale
Identification of fluid components and reliable saturation evaluation remain critical challenges in shale exploration and development. Conventional core experimental methods are often limited by high costs and long cycle times, while laboratory nuclear magnetic resonance (NMR) techniques fail to perform rapid and continuous measurements. In this study, we propose a rock-constrained multilevel Gaussian mixture model (GMM) approach for the quantitative evaluation of multiple fluids in shale using wellsite NMR data. The proposed workflow begins with normalization and thresholding of the T₁–T₂ spectra. A rock-physics constraint is then incorporated to partition the relaxation domain and distinguish movable fluids from bound fluids. GMM clustering is first applied independently within each relaxation region to extract representative fluid signatures. These characteristic signatures are subsequently integrated into a second-stage GMM analysis, enabling robust identification and quantitative evaluation of individual fluid components. Application to representative shale NMR datasets, coupled with multi-state core experiments, demonstrates that the proposed method effectively identifies multiple fluid components, including heavy components, bound oil, and movable oil. Furthermore, the saturation of various hydrocarbon components can be quantitatively predicted, demonstrating strong agreement with Rock-Eval measurements. The proposed approach provides a practical and reliable solution for rapid downhole fluid identification and saturation evaluation in shale reservoirs.
- Open Access
On the Benefit of Dual-domain Denoising in a Self-supervised Low-dose CT Setting
Computed tomography (CT) is routinely used for three-dimensional non-invasive imaging. Numerous data-driven image denoising algorithms were proposed to restore image quality in low-dose acquisitions. However, considerably less research investigates methods already intervening in the raw detector data due to limited access to suitable projection data or correct reconstruction algorithms. In this work, we present an end-to-end trainable CT reconstruction pipeline that contains denoising operators in both the projection and the image domain and that are optimized simultaneously without requiring ground-truth high-dose CT data. Our experiments demonstrate that including an additional projection denoising operator improved the overall denoising performance by 82.4-94.1%/12.5-41.7% (PSNR/SSIM) on abdomen CT and 1.5-2.9%/0.4-0.5% (PSNR/SSIM) on XRM data relative to the low-dose baseline. We make our entire helical CT reconstruction framework publicly available that contains a raw projection rebinning step to render helical projection data suitable for differentiable fan-beam reconstruction operators and end-to-end learning.
Open AccessOn the Benefit of Dual-domain Denoising in a Self-supervised Low-dose CT Setting
Computed tomography (CT) is routinely used for three-dimensional non-invasive imaging. Numerous data-driven image denoising algorithms were proposed to restore image quality in low-dose acquisitions. However, considerably less research investigates methods already intervening in the raw detector data due to limited access to suitable projection data or correct reconstruction algorithms. In this work, we present an end-to-end trainable CT reconstruction pipeline that contains denoising operators in both the projection and the image domain and that are optimized simultaneously without requiring ground-truth high-dose CT data. Our experiments demonstrate that including an additional projection denoising operator improved the overall denoising performance by 82.4-94.1%/12.5-41.7% (PSNR/SSIM) on abdomen CT and 1.5-2.9%/0.4-0.5% (PSNR/SSIM) on XRM data relative to the low-dose baseline. We make our entire helical CT reconstruction framework publicly available that contains a raw projection rebinning step to render helical projection data suitable for differentiable fan-beam reconstruction operators and end-to-end learning.
- Open Access
Trainable Joint Bilateral Filters for Enhanced Prediction Stability in Low-dose CT
Low-dose computed tomography (CT) denoising algorithms aim to enable reduced patient dose in routine CT acquisitions while maintaining high image quality. Recently, deep learning~(DL)-based methods were introduced, outperforming conventional denoising algorithms on this task due to their high model capacity. However, for the transition of DL-based denoising to clinical practice, these data-driven approaches must generalize robustly beyond the seen training data. We, therefore, propose a hybrid denoising approach consisting of a set of trainable joint bilateral filters (JBFs) combined with a convolutional DL-based denoising network to predict the guidance image. Our proposed denoising pipeline combines the high model capacity enabled by DL-based feature extraction with the reliability of the conventional JBF. The pipeline's ability to generalize is demonstrated by training on abdomen CT scans without metal implants and testing on abdomen scans with metal implants as well as on head CT data. When embedding two well-established DL-based denoisers (RED-CNN/QAE) in our pipeline, the denoising performance is improved by $10\,\%$/$82\,\%$ (RMSE) and $3\,\%$/$81\,\%$ (PSNR) in regions containing metal and by $6\,\%$/$78\,\%$ (RMSE) and $2\,\%$/$4\,\%$ (PSNR) on head CT data, compared to the respective vanilla model. Concluding, the proposed trainable JBFs limit the error bound of deep neural networks to facilitate the applicability of DL-based denoisers in low-dose CT pipelines.Open AccessTrainable Joint Bilateral Filters for Enhanced Prediction Stability in Low-dose CT
Low-dose computed tomography (CT) denoising algorithms aim to enable reduced patient dose in routine CT acquisitions while maintaining high image quality. Recently, deep learning~(DL)-based methods were introduced, outperforming conventional denoising algorithms on this task due to their high model capacity. However, for the transition of DL-based denoising to clinical practice, these data-driven approaches must generalize robustly beyond the seen training data. We, therefore, propose a hybrid denoising approach consisting of a set of trainable joint bilateral filters (JBFs) combined with a convolutional DL-based denoising network to predict the guidance image. Our proposed denoising pipeline combines the high model capacity enabled by DL-based feature extraction with the reliability of the conventional JBF. The pipeline's ability to generalize is demonstrated by training on abdomen CT scans without metal implants and testing on abdomen scans with metal implants as well as on head CT data. When embedding two well-established DL-based denoisers (RED-CNN/QAE) in our pipeline, the denoising performance is improved by $10\,\%$/$82\,\%$ (RMSE) and $3\,\%$/$81\,\%$ (PSNR) in regions containing metal and by $6\,\%$/$78\,\%$ (RMSE) and $2\,\%$/$4\,\%$ (PSNR) on head CT data, compared to the respective vanilla model. Concluding, the proposed trainable JBFs limit the error bound of deep neural networks to facilitate the applicability of DL-based denoisers in low-dose CT pipelines. - Open Access
ConFUDA: Contrastive Fewshot Unsupervised Domain Adaptation for Medical Image Segmentation
Unsupervised domain adaptation (UDA) aims to transfer knowledge learned from a labeled source domain to an unlabeled target domain. Contrastive learning (CL) in the context of UDA can help to better separate classes in feature space. However, in image segmentation, the large memory footprint due to the computation of the pixel-wise contrastive loss makes it prohibitive to use. Furthermore, labeled target data is not easily available in medical imaging, and obtaining new samples is not economical. As a result, in this work, we tackle a more challenging UDA task when there are only a few (fewshot) or a single (oneshot) image available from the target domain. We apply a style transfer module to mitigate the scarcity of target samples. Then, to align the source and target features and tackle the memory issue of the traditional contrastive loss, we propose the centroid-based contrastive learning (CCL) and a centroid norm regularizer (CNR) to optimize the contrastive pairs in both direction and magnitude. In addition, we propose multi-partition centroid contrastive learning (MPCCL) to further reduce the variance in the target features. Fewshot evaluation on MS-CMRSeg dataset demonstrates that ConFUDA improves the segmentation performance by 0.34 of the Dice score on the target domain compared with the baseline, and 0.31 Dice score improvement in a more rigorous oneshot setting.
Open AccessConFUDA: Contrastive Fewshot Unsupervised Domain Adaptation for Medical Image Segmentation
Unsupervised domain adaptation (UDA) aims to transfer knowledge learned from a labeled source domain to an unlabeled target domain. Contrastive learning (CL) in the context of UDA can help to better separate classes in feature space. However, in image segmentation, the large memory footprint due to the computation of the pixel-wise contrastive loss makes it prohibitive to use. Furthermore, labeled target data is not easily available in medical imaging, and obtaining new samples is not economical. As a result, in this work, we tackle a more challenging UDA task when there are only a few (fewshot) or a single (oneshot) image available from the target domain. We apply a style transfer module to mitigate the scarcity of target samples. Then, to align the source and target features and tackle the memory issue of the traditional contrastive loss, we propose the centroid-based contrastive learning (CCL) and a centroid norm regularizer (CNR) to optimize the contrastive pairs in both direction and magnitude. In addition, we propose multi-partition centroid contrastive learning (MPCCL) to further reduce the variance in the target features. Fewshot evaluation on MS-CMRSeg dataset demonstrates that ConFUDA improves the segmentation performance by 0.34 of the Dice score on the target domain compared with the baseline, and 0.31 Dice score improvement in a more rigorous oneshot setting.
- Open Access
Few-shot Unsupervised Domain Adaptation for Multi-modal Cardiac Image Segmentation
Unsupervised domain adaptation (UDA) methods intend to reduce the gap between source and target domains by using unlabeled target domain and labeled source domain data, however, in the medical domain, target domain data may not always be easily available, and acquiring new samples is generally time-consuming. This restricts the development of UDA methods for new domains. In this paper, we explore the potential of UDA in a more challenging while realistic scenario where only one unlabeled target patient sample is available. We call it Few-shot Unsupervised Domain adaptation (FUDA). We first generate target-style images from source images and explore diverse target styles from a single target patient with Random Adaptive Instance Normalization (RAIN). Then, a segmentation network is trained in a supervised manner with the generated target images. Our experiments demonstrate that FUDA improves the segmentation performance by 0.33 of Dice score on the target domain compared with the baseline, and it also gives 0.28 of Dice score improvement in a more rigorous one-shot setting. Our code is available at \url{https://github.com/MingxuanGu/Few-shot-UDA}.
Open AccessFew-shot Unsupervised Domain Adaptation for Multi-modal Cardiac Image Segmentation
Unsupervised domain adaptation (UDA) methods intend to reduce the gap between source and target domains by using unlabeled target domain and labeled source domain data, however, in the medical domain, target domain data may not always be easily available, and acquiring new samples is generally time-consuming. This restricts the development of UDA methods for new domains. In this paper, we explore the potential of UDA in a more challenging while realistic scenario where only one unlabeled target patient sample is available. We call it Few-shot Unsupervised Domain adaptation (FUDA). We first generate target-style images from source images and explore diverse target styles from a single target patient with Random Adaptive Instance Normalization (RAIN). Then, a segmentation network is trained in a supervised manner with the generated target images. Our experiments demonstrate that FUDA improves the segmentation performance by 0.33 of Dice score on the target domain compared with the baseline, and it also gives 0.28 of Dice score improvement in a more rigorous one-shot setting. Our code is available at \url{https://github.com/MingxuanGu/Few-shot-UDA}.
- 7 January 2022
- Published by Elsevier in Journal of Natural Gas Science and Engineering
- Vol. 99
A new quantitative evaluation method for fluid constituents with NMR T1-T2 spectra in shale reservoirs
- 7 January 2022
- Published by Elsevier in Journal of Natural Gas Science and Engineering
- Vol. 99
A new quantitative evaluation method for fluid constituents with NMR T1-T2 spectra in shale reservoirs