Brian C. Lovell
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
Research Subjects
Brian C. Lovell has a total of 73 co-authors. Here are the top co-authors by number of shared publications:
- Arnold Wiliem
School of ITEE The University of Queensland 4072 QLD Australia
9 shared publications - Mehrtash Harandi
Monash University, Melbourne, VIC, Australia
5 shared publications - Can Peng
University of Oxford
3 shared publications - Liangchen Liu
School of ITEE The University of Queensland 4072 QLD Australia
3 shared publications - Shaokang Chen
School of ITEE The University of Queensland 4072 QLD Australia
3 shared publications - Gennaro Percannella
University of Salerno
2 shared publications
Latest Publications
- 3 October 2025
- Published by Springer Nature
- p. 272-288
Spectral Distribution Alignment for Enhanced Generalization in Regression
- 3 October 2025
- Published by Springer Nature
- p. 272-288
Spectral Distribution Alignment for Enhanced Generalization in Regression
- 16 March 2024
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 243
Domain-aware triplet loss in domain generalization
- 16 March 2024
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 243
Domain-aware triplet loss in domain generalization
- 28 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 33, 65-72
Uncertainty-Guided Cross-Modality Semi-Supervised Learning for MRI Segmentation
Semi-supervised learning emerges as a viable strategy for addressing the challenge posed by the limited availability of annotated data in medical image segmentation tasks. Nonetheless, current semi-supervised methods for medical image segmentation mainly concentrate on single-modality data. Leveraging multi-modality data effectively to enhance the effectiveness of semi-supervised segmentation methods is a valuable research topic because such data provides additional complementary information. Nevertheless, the majority of current multi-modality semi-supervised segmentation methods are reliant on highly coupled fusion networks, requiring the presence of multi-modality data during both the training and inference phases. It imposes constraints on the deployment of these solutions in clinical practice. To tackle this difficulty, we present UGCM-Semi, a novel multi-modality semi-supervised segmentation framework with an uncertainty-guided distribution calibration (UGDC) student network and a cross-model feature alignment loss. Specifically, we propose the UGDC student network which utilizes the uncertainty modeling technique to quantify and reduce the distribution shift due to labeled and unlabeled data with different modalities. Meanwhile, we add the cross-model feature alignment loss to mitigate the incorrect supervision from the teacher network to the student network caused by the aforementioned distribution shift. Extensive experiments on two public multi-modality MRI datasets demonstrate that UGCN-Semi provides superiority over the state-of-the-art methods.
- 28 November 2023
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 33, 65-72
Uncertainty-Guided Cross-Modality Semi-Supervised Learning for MRI Segmentation
Semi-supervised learning emerges as a viable strategy for addressing the challenge posed by the limited availability of annotated data in medical image segmentation tasks. Nonetheless, current semi-supervised methods for medical image segmentation mainly concentrate on single-modality data. Leveraging multi-modality data effectively to enhance the effectiveness of semi-supervised segmentation methods is a valuable research topic because such data provides additional complementary information. Nevertheless, the majority of current multi-modality semi-supervised segmentation methods are reliant on highly coupled fusion networks, requiring the presence of multi-modality data during both the training and inference phases. It imposes constraints on the deployment of these solutions in clinical practice. To tackle this difficulty, we present UGCM-Semi, a novel multi-modality semi-supervised segmentation framework with an uncertainty-guided distribution calibration (UGDC) student network and a cross-model feature alignment loss. Specifically, we propose the UGDC student network which utilizes the uncertainty modeling technique to quantify and reduce the distribution shift due to labeled and unlabeled data with different modalities. Meanwhile, we add the cross-model feature alignment loss to mitigate the incorrect supervision from the teacher network to the student network caused by the aforementioned distribution shift. Extensive experiments on two public multi-modality MRI datasets demonstrate that UGCN-Semi provides superiority over the state-of-the-art methods.
- Open Access
Domain-aware Triplet loss in Domain Generalization
Despite much progress being made in the field of object recognition with the advances of deep learning, there are still several factors negatively affecting the performance of deep learning models. Domain shift is one of these factors and is caused by discrepancies in the distributions of the testing and training data. In this paper, we focus on the problem of compact feature clustering in domain generalization to help optimize the embedding space from multi-domain data. We design a domainaware triplet loss for domain generalization to help the model to not only cluster similar semantic features, but also to disperse features arising from the domain. Unlike previous methods focusing on distribution alignment, our algorithm is designed to disperse domain information in the embedding space. The basic idea is motivated based on the assumption that embedding features can be clustered based on domain information, which is mathematically and empirically supported in this paper. In addition, during our exploration of feature clustering in domain generalization, we note that factors affecting the convergence of metric learning loss in domain generalization are more important than the pre-defined domains. To solve this issue, we utilize two methods to normalize the embedding space, reducing the internal covariate shift of the embedding features. The ablation study demonstrates the effectiveness of our algorithm. Moreover, the experiments on the benchmark datasets, including PACS, VLCS and Office-Home, show that our method outperforms related methods focusing on domain discrepancy. In particular, our results on RegnetY-16 are significantly better than state-of-the-art methods on the benchmark datasets. Our code will be released at https://github.com/workerbcd/DCT
Open AccessDomain-aware Triplet loss in Domain Generalization
Despite much progress being made in the field of object recognition with the advances of deep learning, there are still several factors negatively affecting the performance of deep learning models. Domain shift is one of these factors and is caused by discrepancies in the distributions of the testing and training data. In this paper, we focus on the problem of compact feature clustering in domain generalization to help optimize the embedding space from multi-domain data. We design a domainaware triplet loss for domain generalization to help the model to not only cluster similar semantic features, but also to disperse features arising from the domain. Unlike previous methods focusing on distribution alignment, our algorithm is designed to disperse domain information in the embedding space. The basic idea is motivated based on the assumption that embedding features can be clustered based on domain information, which is mathematically and empirically supported in this paper. In addition, during our exploration of feature clustering in domain generalization, we note that factors affecting the convergence of metric learning loss in domain generalization are more important than the pre-defined domains. To solve this issue, we utilize two methods to normalize the embedding space, reducing the internal covariate shift of the embedding features. The ablation study demonstrates the effectiveness of our algorithm. Moreover, the experiments on the benchmark datasets, including PACS, VLCS and Office-Home, show that our method outperforms related methods focusing on domain discrepancy. In particular, our results on RegnetY-16 are significantly better than state-of-the-art methods on the benchmark datasets. Our code will be released at https://github.com/workerbcd/DCT
- 20 October 2022
- Published by Springer Nature
- p. 382-397
Few-Shot Class-Incremental Learning from an Open-Set Perspective
- 20 October 2022
- Published by Springer Nature
- p. 382-397
Few-Shot Class-Incremental Learning from an Open-Set Perspective
- 1 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-7
Deep Adaptive Few Example Learning for Microscopy Image Cell Counting
Deep learning networks have demonstrated robust performance for cell counting in medical images. However, obtaining promising results requires large amount of annotated data for supervised training, which is labor-intensive. To address this problem, we propose a novel adaptive few example counting network that aims at localizing cells in microscopy images with only a few annotated examples. We incorporate a few-shot learning phase to transfer information of features from general training images to novel medical images. We also present a training strategy that takes both novel images with their few instances as the input to predict the final results. Furthermore, we introduce an adaptation stage during test time which takes a query medical image as well as its corresponding few object instances to boost the network performance. As the task is firstly established, we further create three few-shot cell counting datasets by adding cell exemplars on top of the public medical images and set them as the baseline. Extensive experiments on the three microscopy datasets show that our framework outperforms the other baseline approaches.
- 1 November 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1-7
Deep Adaptive Few Example Learning for Microscopy Image Cell Counting
Deep learning networks have demonstrated robust performance for cell counting in medical images. However, obtaining promising results requires large amount of annotated data for supervised training, which is labor-intensive. To address this problem, we propose a novel adaptive few example counting network that aims at localizing cells in microscopy images with only a few annotated examples. We incorporate a few-shot learning phase to transfer information of features from general training images to novel medical images. We also present a training strategy that takes both novel images with their few instances as the input to predict the final results. Furthermore, we introduce an adaptation stage during test time which takes a query medical image as well as its corresponding few object instances to boost the network performance. As the task is firstly established, we further create three few-shot cell counting datasets by adding cell exemplars on top of the public medical images and set them as the baseline. Extensive experiments on the three microscopy datasets show that our framework outperforms the other baseline approaches.
- 1 September 2021
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 210
SID: Incremental learning for anchor-free object detection via Selective and Inter-related Distillation
- 1 September 2021
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 210
SID: Incremental learning for anchor-free object detection via Selective and Inter-related Distillation
- Open Access
- 20 February 2021
- Published by American Geophysical Union (AGU) in Journal of Geophysical Research: Space Physics
- Vol. 126 (2)
Investigating the Development of Distinctive Subauroral Flow Channels During the November 7–8, 2004 Superstorm
We investigate the development of large (∼3,000 m/s) subauroral flows occurring during the November 7–8, 2004 Superstorm by utilizing multisatellite and multiinstrument measurements from the Magnetosphere (M), topside Ionosphere (I), and Thermosphere (T). We present eight scenarios (Sc‐1–Sc‐8) depicting the development of large SubAuroral Polarization Streams (SAPS) in the initial phase (Sc‐1–Sc‐3) and recovery phase (Sc‐5), and structured SAPS (Sc‐4, Sc‐6–Sc‐7) and Double‐peak SubAuroral Ion Drifts (DSAID) in the recovery phase (Sc‐8). Magnetospheric observations reveal dayside magnetopause reconnection increasing both the magnetospheric convection electric field and the large‐scale Field‐Aligned Currents (FACs), long‐lasting (∼5 h) secondary dipolarizations during the initial phase, and a series of dipolarization events during the recovery phase. Ionospheric observations reveal the development of these large subauroral flows (i) during forward and reverse plasma convections—when the topside ionosphere received earthward energy from the magnetosphere and when the diversion of cross‐tail currents in the stretched magnetotail activated the Substorm Current Wedge (SCW)—and (ii) in a short‐circuited system that acted as a voltage generator. Thermospheric observations reveal earthward energy deposition from the ionosphere that drove wind surges—implying the presence of Atmospheric Gravity Waves (AGWs) and Traveling Ionospheric Disturbances (TIDs)—propagating equatorward/poleward and passing through the trough region, and further enhancing the subauroral flows by deepening the trough. From these results we conclude that the strong M‐I‐T coupling occurring under superstorm conditions played a significant role in the enhancement of the subauroral flows investigated.
- 20 February 2021
- Published by American Geophysical Union (AGU) in Journal of Geophysical Research: Space Physics
- Vol. 126 (2)
Open AccessInvestigating the Development of Distinctive Subauroral Flow Channels During the November 7–8, 2004 Superstorm
We investigate the development of large (∼3,000 m/s) subauroral flows occurring during the November 7–8, 2004 Superstorm by utilizing multisatellite and multiinstrument measurements from the Magnetosphere (M), topside Ionosphere (I), and Thermosphere (T). We present eight scenarios (Sc‐1–Sc‐8) depicting the development of large SubAuroral Polarization Streams (SAPS) in the initial phase (Sc‐1–Sc‐3) and recovery phase (Sc‐5), and structured SAPS (Sc‐4, Sc‐6–Sc‐7) and Double‐peak SubAuroral Ion Drifts (DSAID) in the recovery phase (Sc‐8). Magnetospheric observations reveal dayside magnetopause reconnection increasing both the magnetospheric convection electric field and the large‐scale Field‐Aligned Currents (FACs), long‐lasting (∼5 h) secondary dipolarizations during the initial phase, and a series of dipolarization events during the recovery phase. Ionospheric observations reveal the development of these large subauroral flows (i) during forward and reverse plasma convections—when the topside ionosphere received earthward energy from the magnetosphere and when the diversion of cross‐tail currents in the stretched magnetotail activated the Substorm Current Wedge (SCW)—and (ii) in a short‐circuited system that acted as a voltage generator. Thermospheric observations reveal earthward energy deposition from the ionosphere that drove wind surges—implying the presence of Atmospheric Gravity Waves (AGWs) and Traveling Ionospheric Disturbances (TIDs)—propagating equatorward/poleward and passing through the trough region, and further enhancing the subauroral flows by deepening the trough. From these results we conclude that the strong M‐I‐T coupling occurring under superstorm conditions played a significant role in the enhancement of the subauroral flows investigated.
- Open Access
Unsupervised Domain Adaptive Object Detection using Forward-Backward Cyclic Adaptation
We present a novel approach to perform the unsupervised domain adaptation for object detection through forward-backward cyclic (FBC) training. Recent adversarial training based domain adaptation methods have shown their effectiveness on minimizing domain discrepancy via marginal feature distributions alignment. However, aligning the marginal feature distributions does not guarantee the alignment of class conditional distributions. This limitation is more evident when adapting object detectors as the domain discrepancy is larger compared to the image classification task, e.g. various number of objects exist in one image and the majority of content in an image is the background. This motivates us to learn domain invariance for category level semantics via gradient alignment. Intuitively, if the gradients of two domains point in similar directions, then the learning of one domain can improve that of another domain. To achieve gradient alignment, we propose Forward-Backward Cyclic Adaptation, which iteratively computes adaptation from source to target via backward hopping and from target to source via forward passing. In addition, we align low-level features for adapting holistic color/texture via adversarial training. However, the detector performs well on both domains is not ideal for target domain. As such, in each cycle, domain diversity is enforced by maximum entropy regularization on the source domain to penalize confident source-specific learning and minimum entropy regularization on target domain to intrigue target-specific learning. Theoretical analysis of the training process is provided, and extensive experiments on challenging cross-domain object detection datasets have shown the superiority of our approach over the state-of-the-art.
Open AccessUnsupervised Domain Adaptive Object Detection using Forward-Backward Cyclic Adaptation
We present a novel approach to perform the unsupervised domain adaptation for object detection through forward-backward cyclic (FBC) training. Recent adversarial training based domain adaptation methods have shown their effectiveness on minimizing domain discrepancy via marginal feature distributions alignment. However, aligning the marginal feature distributions does not guarantee the alignment of class conditional distributions. This limitation is more evident when adapting object detectors as the domain discrepancy is larger compared to the image classification task, e.g. various number of objects exist in one image and the majority of content in an image is the background. This motivates us to learn domain invariance for category level semantics via gradient alignment. Intuitively, if the gradients of two domains point in similar directions, then the learning of one domain can improve that of another domain. To achieve gradient alignment, we propose Forward-Backward Cyclic Adaptation, which iteratively computes adaptation from source to target via backward hopping and from target to source via forward passing. In addition, we align low-level features for adapting holistic color/texture via adversarial training. However, the detector performs well on both domains is not ideal for target domain. As such, in each cycle, domain diversity is enforced by maximum entropy regularization on the source domain to penalize confident source-specific learning and minimum entropy regularization on target domain to intrigue target-specific learning. Theoretical analysis of the training process is provided, and extensive experiments on challenging cross-domain object detection datasets have shown the superiority of our approach over the state-of-the-art.
- 1 January 2020
- Published by University of Queensland Library
Liver Kidney Stomach Immunofluorescence Dataset
The University of Queensland's institutional repository, UQ eSpace, aims to create global visibility and accessibility of UQ’s scholarly research.
- 1 January 2020
- Published by University of Queensland Library
Liver Kidney Stomach Immunofluorescence Dataset
The University of Queensland's institutional repository, UQ eSpace, aims to create global visibility and accessibility of UQ’s scholarly research.