Axel De Nardin
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Axel De Nardin has a total of 45 co-authors. Here are the top co-authors by number of shared publications:
- Gian Luca Foresti
Department of Mathematics, Computer Science and Physics, University of Udine, Via delle Scienze 206, Udine 33100, Italy
19 shared publications - Silvia Zottin
University of Udine
15 shared publications - Claudio Piciarelli
University of Udine
12 shared publications - Emanuela Colombi
University of Udine
5 shared publications - Hussain Ahmad Madni
Department of Mathematics, Computer Science and Physics, University of Udine, Via delle Scienze 206, Udine 33100, Italy
3 shared publications - Filippo Pavan
Department of Humanities and Cultural Heritage, University of Udine, Udine, Italy
2 shared publications
Latest Publications
- 31 August 2026
- Published by World Scientific Pub Co Pte Ltd in International Journal of Neural Systems
- Vol. 37 (03)
Semantic-Aware Semi-Supervised Systematic Segmentation for Brain Magnetic Resonance Image
Accurate segmentation of brain MRI images is essential for early diagnosis and prevention of brain tumors and strokes. However, it is challenging due to data scarcity, domain shift, complexity, and blurred edges. Recent methods use mean-teacher model learning for the semantic details of an individual image based on the efficiency of teacher architecture. However, to enhance the segmentation in brain MRI, the paper introduces SemS4, a semi-supervised method on the basis of cross-image architecture. SemS4 uses a single architecture integrating an Edge Prototype Attention (EPA) and a foreground prototype attention (FPA). More specifically, EPA uses edge prototype integrated with adaptive edge container to improve the edge features for which global edge features are stored to produce edge prototype. Moreover, FPA stabilizes the segmentation by transferring complementary foreground information. Furthermore, the proposed method uses Pixel Affinity Loss (PAL) for missing contextual correlation during supervision to enhance segmentation performance on edges. The performance of the proposed approach in experimental results for two brain MRI benchmarks, LGG and BRISC, validates the superiority over existing methods under different partitioning settings.
- 31 August 2026
- Published by World Scientific Pub Co Pte Ltd in International Journal of Neural Systems
- Vol. 37 (03)
Semantic-Aware Semi-Supervised Systematic Segmentation for Brain Magnetic Resonance Image
Accurate segmentation of brain MRI images is essential for early diagnosis and prevention of brain tumors and strokes. However, it is challenging due to data scarcity, domain shift, complexity, and blurred edges. Recent methods use mean-teacher model learning for the semantic details of an individual image based on the efficiency of teacher architecture. However, to enhance the segmentation in brain MRI, the paper introduces SemS4, a semi-supervised method on the basis of cross-image architecture. SemS4 uses a single architecture integrating an Edge Prototype Attention (EPA) and a foreground prototype attention (FPA). More specifically, EPA uses edge prototype integrated with adaptive edge container to improve the edge features for which global edge features are stored to produce edge prototype. Moreover, FPA stabilizes the segmentation by transferring complementary foreground information. Furthermore, the proposed method uses Pixel Affinity Loss (PAL) for missing contextual correlation during supervision to enhance segmentation performance on edges. The performance of the proposed approach in experimental results for two brain MRI benchmarks, LGG and BRISC, validates the superiority over existing methods under different partitioning settings.
- 26 August 2026
- Published by Springer Nature
- p. 558-574
GRaF-Net: a Multi-branch Gated Residual Architecture for Floor Plan Semantic Segmentation
- 26 August 2026
- Published by Springer Nature
- p. 558-574
GRaF-Net: a Multi-branch Gated Residual Architecture for Floor Plan Semantic Segmentation
- 24 August 2026
- Published by Springer Nature
- p. 370-386
Bridging the Gaps: Learning to Estimate Missing Text in Fragmentary Greek Inscriptions
- 24 August 2026
- Published by Springer Nature
- p. 370-386
Bridging the Gaps: Learning to Estimate Missing Text in Fragmentary Greek Inscriptions
- Open Access
- 18 April 2026
- Published by Springer Nature in International Journal on Document Analysis and Recognition (IJDAR)
- Vol. 29 (2) , 647-658
U-DIADS-TL: a novel dataset for text line segmentation in historical manuscripts
Text line segmentation in historical documents remains a significant challenge due to degraded manuscripts, complex layouts, and diverse handwriting styles. Developing robust computational methods is hindered by the scarcity of high-quality ground truth annotations, which require expert knowledge and are time-intensive to produce. Few-shot learning has emerged as a promising solution by enabling model training with minimal annotated data, yet its application to historical document analysis is still largely unexplored. To address this limitation, we introduce U-DIADS-TL (Uniud - Document Image Analysis DataSet - Text Line), a dataset specifically designed for text line segmentation in ancient manuscripts. U-DIADS-TL provides noise-free annotations with non-overlapping text elements and accommodates diverse document structures, including multi-column layouts. To encourage few-shot learning approaches, we offer only three training images, allowing researchers to develop segmentation models that can generalize from limited supervision. Our dataset serves as a critical bridge between deep learning and historical document analysis, fostering the creation of efficient, adaptable segmentation models for real-world applications.
- 18 April 2026
- Published by Springer Nature in International Journal on Document Analysis and Recognition (IJDAR)
- Vol. 29 (2) , 647-658
Open AccessU-DIADS-TL: a novel dataset for text line segmentation in historical manuscripts
Text line segmentation in historical documents remains a significant challenge due to degraded manuscripts, complex layouts, and diverse handwriting styles. Developing robust computational methods is hindered by the scarcity of high-quality ground truth annotations, which require expert knowledge and are time-intensive to produce. Few-shot learning has emerged as a promising solution by enabling model training with minimal annotated data, yet its application to historical document analysis is still largely unexplored. To address this limitation, we introduce U-DIADS-TL (Uniud - Document Image Analysis DataSet - Text Line), a dataset specifically designed for text line segmentation in ancient manuscripts. U-DIADS-TL provides noise-free annotations with non-overlapping text elements and accommodates diverse document structures, including multi-column layouts. To encourage few-shot learning approaches, we offer only three training images, allowing researchers to develop segmentation models that can generalize from limited supervision. Our dataset serves as a critical bridge between deep learning and historical document analysis, fostering the creation of efficient, adaptable segmentation models for real-world applications.
- Open Access
- 30 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 49032-49054
Mitigating Data Scarcity in Cancer Classification With Synthetic Data
Clinical datasets are often limited in availability and subject to strict privacy regulations, posing significant challenges for the development of accurate classification models. Synthetic data generation offers a promising alternative, enabling the augmentation of training datasets while preserving patient privacy. However, generating high-fidelity synthetic images that effectively support model training remains a significant challenge. In this paper, we focus on the classification of colorectal and lung carcinoma as representative tasks in cancer clinical diagnostics. We utilize a stable diffusion model enhanced with Low-Rank Adaptation (LoRA) weights to generate synthetic images from a limited number of real images used for fine-tuning. This method results in a performance improvement of the DeiT-L (Data-efficient Image Transformer–Large) and CLIP (Contrastive Language–Image Pretraining) models for colon and lung datasets, respectively, when trained on excessive synthetic data and a few real samples. Synthetic data closely mirrors the real samples, mitigating the issues of data scarcity and privacy, while enhancing model generalization. These findings support the use of synthetic data as a viable tool in cancer disease classification and demonstrate its potential to strengthen deep learning applications in clinical diagnostics and medical research. Experimental results on colon and lung histopathology datasets demonstrate that augmenting limited real data with diffusion-generated synthetic images consistently improves classification accuracy and generalization across multiple deep learning architectures, particularly in low-data (few-shot) regimes. Notably, the hybrid configuration combining a small number of real samples with synthetic data yields the most practically significant performance gains, highlighting its relevance for real-world clinical settings where annotated data are scarce. Code is available at: https://github.com/h-ahmad/rare_disease_ classification.
- 30 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 49032-49054
Open AccessMitigating Data Scarcity in Cancer Classification With Synthetic Data
Clinical datasets are often limited in availability and subject to strict privacy regulations, posing significant challenges for the development of accurate classification models. Synthetic data generation offers a promising alternative, enabling the augmentation of training datasets while preserving patient privacy. However, generating high-fidelity synthetic images that effectively support model training remains a significant challenge. In this paper, we focus on the classification of colorectal and lung carcinoma as representative tasks in cancer clinical diagnostics. We utilize a stable diffusion model enhanced with Low-Rank Adaptation (LoRA) weights to generate synthetic images from a limited number of real images used for fine-tuning. This method results in a performance improvement of the DeiT-L (Data-efficient Image Transformer–Large) and CLIP (Contrastive Language–Image Pretraining) models for colon and lung datasets, respectively, when trained on excessive synthetic data and a few real samples. Synthetic data closely mirrors the real samples, mitigating the issues of data scarcity and privacy, while enhancing model generalization. These findings support the use of synthetic data as a viable tool in cancer disease classification and demonstrate its potential to strengthen deep learning applications in clinical diagnostics and medical research. Experimental results on colon and lung histopathology datasets demonstrate that augmenting limited real data with diffusion-generated synthetic images consistently improves classification accuracy and generalization across multiple deep learning architectures, particularly in low-data (few-shot) regimes. Notably, the hybrid configuration combining a small number of real samples with synthetic data yields the most practically significant performance gains, highlighting its relevance for real-world clinical settings where annotated data are scarce. Code is available at: https://github.com/h-ahmad/rare_disease_ classification.
- 2 February 2026
- Published by SAGE Publications in Integrated Computer-Aided Engineering
- Vol. 33 (2) , 182-199
Deep acoustic learning on unmanned aerial vehicles for real-time human and drone detection
In autonomous systems and robotics, acoustic signals provide valuable information for tasks such as acoustic source localization and recognition (LR), particularly in environments where visual sensing is limited. This paper investigates two unmanned aerial vehicles (UAVs)-based real-world scenarios that leverage acoustic scene awareness: (1) localization and recognition of human speech for search-and-rescue missions, and (2) detection and classification of other UAVs for counter-drone applications. To address these tasks, we design two deep learning models based on convolutional neural networks (CNNs) and a feature-based approach. These models process acoustic signals captured by two types of microphone arrays mounted on UAVs: a 4-microphone linear array and a 19-microphone spherical array. Each model performs direction of arrival (DOA) estimation and source classification under challenging ego-noise conditions using real-world datasets recorded in controlled experimental setups. We evaluate the models across different signal-to-ego-noise ratios and training configurations. Results show robust performance in both localization and recognition tasks, with approximately 6 degrees mean error and 7 degrees root mean square error (RMSE) for DOA estimations in the human speech scenario with multi-speaker classification accuracy till 0.95, and 3-5 degrees mean error and 7-11 degrees RMSE for DOA estimations in the UAV sound scenario with multi-UAV classification accuracy till 0.98. This demonstrates the potential of deep acoustic learning for UAV-based scene understanding in complex operational environments.
- 2 February 2026
- Published by SAGE Publications in Integrated Computer-Aided Engineering
- Vol. 33 (2) , 182-199
Deep acoustic learning on unmanned aerial vehicles for real-time human and drone detection
In autonomous systems and robotics, acoustic signals provide valuable information for tasks such as acoustic source localization and recognition (LR), particularly in environments where visual sensing is limited. This paper investigates two unmanned aerial vehicles (UAVs)-based real-world scenarios that leverage acoustic scene awareness: (1) localization and recognition of human speech for search-and-rescue missions, and (2) detection and classification of other UAVs for counter-drone applications. To address these tasks, we design two deep learning models based on convolutional neural networks (CNNs) and a feature-based approach. These models process acoustic signals captured by two types of microphone arrays mounted on UAVs: a 4-microphone linear array and a 19-microphone spherical array. Each model performs direction of arrival (DOA) estimation and source classification under challenging ego-noise conditions using real-world datasets recorded in controlled experimental setups. We evaluate the models across different signal-to-ego-noise ratios and training configurations. Results show robust performance in both localization and recognition tasks, with approximately 6 degrees mean error and 7 degrees root mean square error (RMSE) for DOA estimations in the human speech scenario with multi-speaker classification accuracy till 0.95, and 3-5 degrees mean error and 7-11 degrees RMSE for DOA estimations in the UAV sound scenario with multi-UAV classification accuracy till 0.98. This demonstrates the potential of deep acoustic learning for UAV-based scene understanding in complex operational environments.
- 2 January 2026
- Published by Springer Nature
- p. 416-428
Segmenting Anything in Architecture: A Tailored Approach to Segmenting Floor Plan Documents
- 2 January 2026
- Published by Springer Nature
- p. 416-428
Segmenting Anything in Architecture: A Tailored Approach to Segmenting Floor Plan Documents
- Open Access
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 1
Improving Industrial Anomaly Clustering via Heatmap-Based Multi-Scale Views
- 1 January 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 14, 1
Open AccessImproving Industrial Anomaly Clustering via Heatmap-Based Multi-Scale Views
- Open Access
- 7 November 2025
- Published by Elsevier in Pattern Recognition Letters
- Vol. 199, 113-119
FsBAD: Data-efficient feature reconstruction for few-shot brain anomaly detection
- 7 November 2025
- Published by Elsevier in Pattern Recognition Letters
- Vol. 199, 113-119
Open AccessFsBAD: Data-efficient feature reconstruction for few-shot brain anomaly detection
- 16 September 2025
- Published by Springer Nature
- p. 586-602
ICDAR 2025 Competition on FEw-Shot Text Line Segmentation of Ancient Handwritten Documents (FEST)
- 16 September 2025
- Published by Springer Nature
- p. 586-602
ICDAR 2025 Competition on FEw-Shot Text Line Segmentation of Ancient Handwritten Documents (FEST)