Miguel Angel Garcia
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
Research Interests
Miguel Angel Garcia has a total of 14 co-authors. Here are the top co-authors by number of shared publications:
- Hatem A. Rashwan
DEIM, Rovira i Virgili University, 43007 Tarragona, Spain
5 shared publications - Domenec Puig
Department of Computer Engineering and Mathematics, Universitat Rovira i Virgili, 43007 Tarragona, Spain
4 shared publications - Hussein Hasan
Department of Computer Engineering and Mathematics, University Rovira i Virgili, Tarragona, Spain
3 shared publications Domenec Puig
2 shared publications- Daniel Padilla Carrasco
Department of Computer Engineering and Mathematics, Universitat Rovira i Virgili, Tarragona, Spain
1 shared publication Domènec Puig
Department of Computer Engineering and Mathematics, Universitat Rovira i Virgili, Tarragona, Spain
1 shared publication
Latest Publications
- Open Access
- 1 October 2026
- Published by IOS Press
- p. 324-327
Quaternion-Aware Rotation Modeling for Camera Localization
Camera localization aims to estimate camera position and orientation from visual data. While recent deep learning methods formulate pose estimation as a direct regression problem, rotation is typically represented using quaternions yet estimated via standard real-valued transformations, creating a structural mismatch between learned features in Rd and the rotation manifold S3. We introduce a quaternion-aware rotation modeling approach that incorporates Hamilton-consistent interactions within the feature transformation process, replacing the conventional rotation head with a structured projection stage that explicitly couples quaternion components. Built upon CoHAtNet, our method achieves consistent improvements in median rotation error across all seven scenes of the 7-Scenes dataset, reducing the average rotation error from 0.74° to 0.67°, while maintaining competitive translation accuracy.
- 1 October 2026
- Published by IOS Press
- p. 324-327
Open AccessQuaternion-Aware Rotation Modeling for Camera Localization
Camera localization aims to estimate camera position and orientation from visual data. While recent deep learning methods formulate pose estimation as a direct regression problem, rotation is typically represented using quaternions yet estimated via standard real-valued transformations, creating a structural mismatch between learned features in Rd and the rotation manifold S3. We introduce a quaternion-aware rotation modeling approach that incorporates Hamilton-consistent interactions within the feature transformation process, replacing the conventional rotation head with a structured projection stage that explicitly couples quaternion components. Built upon CoHAtNet, our method achieves consistent improvements in median rotation error across all seven scenes of the 7-Scenes dataset, reducing the average rotation error from 0.74° to 0.67°, while maintaining competitive translation accuracy.
- Open Access
- 6 August 2026
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 271
CoAtXNet: A dual-stream hybrid transformer based on relative cross-attention for end-to-end camera localization from RGB-D images
- 6 August 2026
- Published by Elsevier in Computer Vision and Image Understanding
- Vol. 271
Open AccessCoAtXNet: A dual-stream hybrid transformer based on relative cross-attention for end-to-end camera localization from RGB-D images
- Open Access
- 22 September 2025
- Published by IOS Press
- p. 253-265
Efficient Lung Cancer Classification Using Region-of-Interest-Based Deep Learning
Lung cancer classification using deep learning (DL) has shown promising results, yet many existing approaches rely on processing entire lung images, leading to increased computational cost and potential distractions from irrelevant regions. In this work, we propose a focused and efficient DL-based classification framework that leverages only the regions of interest (ROIs) identified by bound- ing boxes. By isolating tumor areas prior to classification, our method directs the model’s attention solely to clinically relevant features, reducing noise and improving diagnostic accuracy. Unlike segmentation-based or full-image classifiers, our approach eliminates the need for pixel-level annotations and avoids the overhead of complex pre-processing pipelines. The proposed framework inputs cropped tumor regions and outputs the corresponding lung cancer class, offering a lightweight and effective alternative for real-world deployment. Experimental results confirm that targeting ROIs significantly enhances classification performance while streamlining the inference process, supporting its applicability in clinical workflows.
- 22 September 2025
- Published by IOS Press
- p. 253-265
Open AccessEfficient Lung Cancer Classification Using Region-of-Interest-Based Deep Learning
Lung cancer classification using deep learning (DL) has shown promising results, yet many existing approaches rely on processing entire lung images, leading to increased computational cost and potential distractions from irrelevant regions. In this work, we propose a focused and efficient DL-based classification framework that leverages only the regions of interest (ROIs) identified by bound- ing boxes. By isolating tumor areas prior to classification, our method directs the model’s attention solely to clinically relevant features, reducing noise and improving diagnostic accuracy. Unlike segmentation-based or full-image classifiers, our approach eliminates the need for pixel-level annotations and avoids the overhead of complex pre-processing pipelines. The proposed framework inputs cropped tumor regions and outputs the corresponding lung cancer class, offering a lightweight and effective alternative for real-world deployment. Experimental results confirm that targeting ROIs significantly enhances classification performance while streamlining the inference process, supporting its applicability in clinical workflows.
- Open Access
- 26 July 2025
- Published by Elsevier in Image and Vision Computing
- Vol. 162
CoHAtNet: An integrated convolutional-transformer architecture with hybrid self-attention for end-to-end camera localization
- 26 July 2025
- Published by Elsevier in Image and Vision Computing
- Vol. 162
Open AccessCoHAtNet: An integrated convolutional-transformer architecture with hybrid self-attention for end-to-end camera localization
- Open Access
- 4 August 2024
- Published by Wiley in Advanced Intelligent Systems
- Vol. 6 (11)
Dimension Reduction of Multidimensional Structured and Unstructured Datasets through Ensemble Learning of Neural Embeddings
Dimension reduction aims to project a high‐dimensional dataset into a low‐dimensional space. It tries to preserve the topological relationships among the original data points and/or induce clusters. NetDRm, an online dimensionality reduction method based on neural ensemble learning that integrates different dimension reduction methods in a synergistic way, is introduced. NetDRm is designed for datasets of multidimensional points that can be either structured (e.g., images) or unstructured (e.g., point clouds, tabular data). It starts by training a collection of deep residual encoders that learn the embeddings induced by multiple dimension reduction methods applied to the input dataset. Subsequently, a dense neural network integrates the generated encoders by emphasizing topological preservation or cluster induction. Experiments conducted on widely used multidimensional datasets (point‐cloud manifolds, image datasets, tabular record datasets) show that the proposed method yields better results in terms of topological preservation ( curves), cluster induction ( V measure), and classification accuracy than the most relevant dimension reduction methods.
- 4 August 2024
- Published by Wiley in Advanced Intelligent Systems
- Vol. 6 (11)
Open AccessDimension Reduction of Multidimensional Structured and Unstructured Datasets through Ensemble Learning of Neural Embeddings
Dimension reduction aims to project a high‐dimensional dataset into a low‐dimensional space. It tries to preserve the topological relationships among the original data points and/or induce clusters. NetDRm, an online dimensionality reduction method based on neural ensemble learning that integrates different dimension reduction methods in a synergistic way, is introduced. NetDRm is designed for datasets of multidimensional points that can be either structured (e.g., images) or unstructured (e.g., point clouds, tabular data). It starts by training a collection of deep residual encoders that learn the embeddings induced by multiple dimension reduction methods applied to the input dataset. Subsequently, a dense neural network integrates the generated encoders by emphasizing topological preservation or cluster induction. Experiments conducted on widely used multidimensional datasets (point‐cloud manifolds, image datasets, tabular record datasets) show that the proposed method yields better results in terms of topological preservation ( curves), cluster induction ( V measure), and classification accuracy than the most relevant dimension reduction methods.
- Open Access
- 22 December 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 22430-22440
T-YOLO: Tiny Vehicle Detection Based on YOLO and Multi-Scale Convolutional Neural Networks
To solve real-life problems for different smart city applications, using deep Neural Network, such as parking occupancy detection, requires fine-tuning of these networks. For large parking, it is desirable to use a cenital-plane camera located at a high distance that allows the monitoring of the entire parking space or a large parking area with only one camera. Today’s most popular object detection models, such as YOLO, achieve good precision scores at real-time speed. However, if we use our own data different from that of the general-purpose datasets, such as COCO and ImageNet, we have a large margin for improvisation. In this paper, we propose a modified, yet lightweight, deep object detection model based on the YOLO-v5 architecture. The proposed model can detect large, small, and tiny objects. Specifically, we propose the use of a multi-scale mechanism to learn deep discriminative feature representations at different scales and automatically determine the most suitable scales for detecting objects in a scene (i.e., in our case vehicles). The proposed multi-scale module reduces the number of trainable parameters compared to the original YOLO-v5 architecture. The experimental results also demonstrate that precision is improved by a large margin. In fact, as shown in the experiments, the results show a small reduction from 7.28 million parameters of the YOLO-v5-S profile to 7.26 million parameters in our model. In addition, we reduced the detection speed by inferring 30 fps compared to the YOLO-v5-L/X profiles. In addition, the tiny vehicle detection performance was significantly improved by 33% compared to the YOLO-v5-X profile.
- 22 December 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 11, 22430-22440
Open AccessT-YOLO: Tiny Vehicle Detection Based on YOLO and Multi-Scale Convolutional Neural Networks
To solve real-life problems for different smart city applications, using deep Neural Network, such as parking occupancy detection, requires fine-tuning of these networks. For large parking, it is desirable to use a cenital-plane camera located at a high distance that allows the monitoring of the entire parking space or a large parking area with only one camera. Today’s most popular object detection models, such as YOLO, achieve good precision scores at real-time speed. However, if we use our own data different from that of the general-purpose datasets, such as COCO and ImageNet, we have a large margin for improvisation. In this paper, we propose a modified, yet lightweight, deep object detection model based on the YOLO-v5 architecture. The proposed model can detect large, small, and tiny objects. Specifically, we propose the use of a multi-scale mechanism to learn deep discriminative feature representations at different scales and automatically determine the most suitable scales for detecting objects in a scene (i.e., in our case vehicles). The proposed multi-scale module reduces the number of trainable parameters compared to the original YOLO-v5 architecture. The experimental results also demonstrate that precision is improved by a large margin. In fact, as shown in the experiments, the results show a small reduction from 7.28 million parameters of the YOLO-v5-S profile to 7.26 million parameters in our model. In addition, we reduced the detection speed by inferring 30 fps compared to the YOLO-v5-L/X profiles. In addition, the tiny vehicle detection performance was significantly improved by 33% compared to the YOLO-v5-X profile.
- 10 February 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 30 (9) , 2947-2958
Adversarial Learning for Depth and Viewpoint Estimation From a Single Image
- 10 February 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Circuits and Systems for Video Technology
- Vol. 30 (9) , 2947-2958
Adversarial Learning for Depth and Viewpoint Estimation From a Single Image
- 1 January 2003
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 2
Video object motion applications focusing on non-planar rotation
In this paper an object-based non-planar rotation estimation for video analysis and video coding is presented. This method is based on a non-planar rotation model which assumes that the moving object as a planar surface. It is also assumed the considered video objects are rigid and have been previously segmented for each key instant considered. The model can be easily adjusted by choosing a suitable region of support (block or arbitrary rigid region). The presented results indicate that the proposed technique is a suitable approach for motion estimation, which can be used for applications such as MPEG-4 block based hybrid coding or MPEG-4 object based video manipulations. Experimental results have been performed on real video sequences.
- 1 January 2003
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 2
Video object motion applications focusing on non-planar rotation
In this paper an object-based non-planar rotation estimation for video analysis and video coding is presented. This method is based on a non-planar rotation model which assumes that the moving object as a planar surface. It is also assumed the considered video objects are rigid and have been previously segmented for each key instant considered. The model can be easily adjusted by choosing a suitable region of support (block or arbitrary rigid region). The presented results indicate that the proposed technique is a suitable approach for motion estimation, which can be used for applications such as MPEG-4 block based hybrid coding or MPEG-4 object based video manipulations. Experimental results have been performed on real video sequences.