Dylan Seychell

Matthias Bartolo, Dylan Seychell, Josef Bajada

With the ever-growing variety of object detection approaches, this study explores a series of experiments that combine reinforcement learning (RL)-based visual attention methods with saliency ranking techniques to investigate transparent and sustainable solutions. By integrating saliency ranking for initial bounding box prediction and subsequently applying RL techniques to refine these predictions through a finite set of actions over multiple time steps, this study aims to enhance RL object detection accuracy. Presented as a series of experiments, this research investigates the use of various image feature extraction methods and explores diverse Deep Q-Network (DQN) architectural variations for deep reinforcement learning-based localisation agent training. Additionally, we focus on optimising the detection pipeline at every step by prioritising lightweight and faster models, while also incorporating the capability to classify detected objects, a feature absent in previous RL approaches. We show that by evaluating the performance of these trained agents using the Pascal VOC 2007 dataset, faster and more optimised models were developed. Notably, the best mean Average Precision (mAP) achieved in this study was 51.4, surpassing benchmarks set by RL-based single object detectors in the literature.

Dylan Seychell, Gabriel Hili, Jonathan Attard et al.

In today`s media landscape, the role of Artificial Intelligence (AI) in shaping societal perspectives and journalistic integrity is becoming increasingly apparent. This paper presents two case studies centred on Malta`s media market featuring technical novelty. Despite its relatively small scale, Malta offers invaluable insights applicable to both similar and broader media contexts. These two projects focus on media monitoring and present tools designed to analyse potential biases in news articles and television news segments. The first project uses Computer Vision and Natural Language Processing techniques to analyse the coherence between images in news articles and their corresponding captions, headlines, and article bodies. The second project employs computer vision techniques to track individuals` on-screen time or visual exposure in news videos, providing queryable data. These initiatives aim to contribute to society by providing both journalists and the public with the means to identify biases. Furthermore, we make these tools accessible to journalists to improve the trustworthiness of media outlets by offering robust tools for detecting and reducing bias.

Matthias Bartolo, Dylan Seychell, Gabriel Hili et al.

This paper investigates the integration of the Learning Using Privileged Information (LUPI) paradigm in object detection to exploit fine-grained, descriptive information available during training but not at inference. We introduce a general, model-agnostic methodology for injecting privileged information-such as bounding box masks, saliency maps, and depth cues-into deep learning-based object detectors through a teacher-student architecture. Experiments are conducted across five state-of-the-art object detection models and multiple public benchmarks, including UAV-based litter detection datasets and Pascal VOC 2012, to assess the impact on accuracy, generalization, and computational efficiency. Our results demonstrate that LUPI-trained students consistently outperform their baseline counterparts, achieving significant boosts in detection accuracy with no increase in inference complexity or model size. Performance improvements are especially marked for medium and large objects, while ablation studies reveal that intermediate weighting of teacher guidance optimally balances learning from privileged and standard inputs. The findings affirm that the LUPI framework provides an effective and practical strategy for advancing object detection systems in both resource-constrained and real-world settings.

Matthew Kenely, Mark Bugeja, Andre Grima et al.

We present a comprehensive methodology for the large-scale photogrammetric documentation of St. John's Co-Cathedral in Valletta, Malta, a UNESCO World Heritage site renowned for its ornate Baroque architecture and Caravaggio masterpieces. Over seven nights of evening-only data collection, we captured 99,000 images using DSLR cameras, drone photography, and LIDAR scanning to create a highly detailed 3D reconstruction comprising 25-30 billion triangles. This paper documents our complete workflow for cultural heritage preservation, addressing the unique challenges of digitizing complex baroque architectural spaces with highly reflective metallic surfaces, dark materials, intricate tapestries, and restricted access. We detail our pipeline from multi-modal data acquisition through processing, including strategic image grading and AI-assisted denoising to address low-light grain, extensive LIDAR point cloud cleanup, hybrid photogrammetric reconstruction using RealityCapture, and mesh subdivision strategies for real-time visualization engines. Our methodology combines automated workflows with necessary manual intervention to handle the scale and complexity of the project, with particular attention to reflective surface challenges characteristic of baroque heritage sites. We also present preliminary experiments with Gaussian splatting as a complementary representation technique. The resulting digital archive serves multiple preservation purposes including disaster recovery documentation, conservation analysis, virtual tourism, and scholarly research. This work provides a detailed, replicable workflow for heritage professionals undertaking similar large-scale architectural documentation projects, addressing the practical challenges of applying photogrammetric methods in complex real-world heritage scenarios.

Matthias Bartolo, Dylan Seychell

As object detection techniques continue to evolve, understanding their relationships with complementary visual tasks becomes crucial for optimising model architectures and computational resources. This paper investigates the correlations between object detection accuracy and two fundamental visual tasks: depth prediction and visual saliency prediction. Through comprehensive experiments using state-of-the-art models (DeepGaze IIE, Depth Anything, DPT-Large, and Itti's model) on COCO and Pascal VOC datasets, we find that visual saliency shows consistently stronger correlations with object detection accuracy (mA$ρ$ up to 0.459 on Pascal VOC) compared to depth prediction (mA$ρ$ up to 0.283). Our analysis reveals significant variations in these correlations across object categories, with larger objects showing correlation values up to three times higher than smaller objects. These findings suggest incorporating visual saliency features into object detection architectures could be more beneficial than depth information, particularly for specific object categories. The observed category-specific variations also provide insights for targeted feature engineering and dataset design improvements, potentially leading to more efficient and accurate object detection systems.

Andrea Filiberto Lucas, Dylan Seychell

The growing volume of video-based news content has heightened the need for transparent and reliable methods to extract on-screen information. Yet the variability of graphical layouts, typographic conventions, and platform-specific design patterns renders manual indexing impractical. This work presents a comprehensive framework for automatically detecting and extracting personal names from broadcast and social-media-native news videos. It introduces a curated and balanced corpus of annotated frames capturing the diversity of contemporary news graphics and proposes an interpretable, modular extraction pipeline designed to operate under deterministic and auditable conditions. The pipeline is evaluated against a contrasting class of generative multimodal methods, revealing a clear trade-off between deterministic auditability and stochastic inference. The underlying detector achieves 95.8% mAP@0.5, demonstrating operationally robust performance for graphical element localisation. While generative systems achieve marginally higher raw accuracy (F1: 84.18% vs 77.08%), they lack the transparent data lineage required for journalistic and analytical contexts. The proposed pipeline delivers balanced precision (79.9%) and recall (74.4%), avoids hallucination, and provides full traceability across each processing stage. Complementary user findings indicate that 59% of respondents report difficulty reading on-screen names in fast-paced broadcasts, underscoring the practical relevance of the task. The results establish a methodologically rigorous and interpretable baseline for hybrid multimodal information extraction in modern news media.

Matthias Bartolo, Konstantinos Makantasis, Dylan Seychell

As litter pollution continues to rise globally, developing automated tools capable of detecting litter effectively remains a significant challenge. This study presents a novel approach that combines, for the first time, privileged information with deep learning object detection to improve litter detection while maintaining model efficiency. We evaluate our method across five widely used object detection models, addressing challenges such as detecting small litter and objects partially obscured by grass or stones. In addition to this, a key contribution of our work can also be attributed to formulating a means of encoding bounding box information as a binary mask, which can be fed to the detection model to refine detection guidance. Through experiments on both within-dataset evaluation on the renowned SODA dataset and cross-dataset evaluation on the BDW and UAVVaste litter detection datasets, we demonstrate consistent performance improvements across all models. Our approach not only bolsters detection accuracy within the training sets but also generalises well to other litter detection contexts. Crucially, these improvements are achieved without increasing model complexity or adding extra layers, ensuring computational efficiency and scalability. Our results suggest that this methodology offers a practical solution for litter detection, balancing accuracy and efficiency in real-world applications.

Jonathan Attard, Dylan Seychell

News videos require efficient content organisation and retrieval systems, but their unstructured nature poses significant challenges for automated processing. This paper presents a comprehensive comparative analysis of image, video, and audio classifiers for automated news video segmentation. This work presents the development and evaluation of multiple deep learning approaches, including ResNet, ViViT, AST, and multimodal architectures, to classify five distinct segment types: advertisements, stories, studio scenes, transitions, and visualisations. Using a custom-annotated dataset of 41 news videos comprising 1,832 scene clips, our experiments demonstrate that image-based classifiers achieve superior performance (84.34\% accuracy) compared to more complex temporal models. Notably, the ResNet architecture outperformed state-of-the-art video classifiers while requiring significantly fewer computational resources. Binary classification models achieved high accuracy for transitions (94.23\%) and advertisements (92.74\%). These findings advance the understanding of effective architectures for news video segmentation and provide practical insights for implementing automated content organisation systems in media applications. These include media archiving, personalised content delivery, and intelligent video search.

Matthew Kenely, Dylan Seychell, Carl James Debono et al.

News outlets' competition for attention in news interfaces has highlighted the need for demographically-aware saliency prediction models. Despite recent advancements in saliency detection applied to user interfaces (UI), existing datasets are limited in size and demographic representation. We present a deep learning framework that enhances the SaRa (Saliency Ranking) model with DeepGaze IIE, improving Salient Object Ranking (SOR) performance by 10.7%. Our framework optimizes three key components: saliency map generation, grid segment scoring, and map normalization. Through a two-fold experiment using eye-tracking (30 participants) and mouse-tracking (375 participants aged 13--70), we analyze attention patterns across demographic groups. Statistical analysis reveals significant age-based variations (p < 0.05, {ε^2} = 0.042), with older users (36--70) engaging more with textual content and younger users (13--35) interacting more with images. Mouse-tracking data closely approximates eye-tracking behavior (sAUC = 0.86) and identifies UI elements that immediately stand out, validating its use in large-scale studies. We conclude that saliency studies should prioritize gathering data from a larger, demographically representative sample and report exact demographic distributions.

Ljubisa Bojic, Dylan Seychell, Milan Cabarkapa

As general-purpose artificial intelligence systems become increasingly integrated into society and are used for information seeking, content generation, problem solving, textual analysis, coding, and running processes, it is crucial to assess their long-term impact on humans. This research explores the sentiment of large language models (LLMs) and humans toward artificial general intelligence (AGI) using a Likert-scale survey. Seven LLMs, including GPT-4 and Bard, were analyzed and compared with sentiment data from three independent human sample populations. Temporal variations in sentiment were also evaluated over three consecutive days. The results show a diversity in sentiment scores among LLMs, ranging from 3.32 to 4.12 out of 5. GPT-4 recorded the most positive sentiment toward AGI, while Bard leaned toward a neutral sentiment. In contrast, the human samples showed a lower average sentiment of 2.97. The analysis outlines potential conflicts of interest and biases in the sentiment formation of LLMs, and indicates that LLMs could subtly influence societal perceptions. To address the need for regulatory oversight and culturally grounded assessments of AI systems, we introduce the Societal AI Alignment and Sentiment Benchmark (SAAS-AI), which leverages multidimensional prompts and empirically validated societal value frameworks to evaluate language model outputs across temporal, model, and multilingual axes. This benchmark is designed to guide policymakers and AI agencies, including within frameworks such as the EU AI Act, by providing robust, actionable insights into AI alignment with human values, public sentiment, and ethical norms at both national and international levels. Future research should further refine the operationalization of the SAAS-AI benchmark and systematically evaluate its effectiveness through comprehensive empirical testing.

Michele La Ferla, Matthew Montebello, Dylan Seychell

The results showed that XAI could indeed be used as a proof of concept to begin discussions on the implementation of assistive AI systems within the clinical community.

Stefan Cassar, Adrian Muscat, Dylan Seychell

Vision and language tasks such as Visual Relation Detection and Visual Question Answering benefit from semantic features that afford proper grounding of language. The 3D depth of objects depicted in 2D images is one such feature. However it is very difficult to obtain accurate depth information without learning the appropriate features, which are scene dependent. The state of the art in this area are complex Neural Network models trained on stereo image data to predict depth per pixel. Fortunately, in some tasks, its only the relative depth between objects that is required. In this paper the extent to which semantic features can predict course relative depth is investigated. The problem is casted as a classification one and geometrical features based on object bounding boxes, object labels and scene attributes are computed and used as inputs to pattern recognition models to predict relative depth. i.e behind, in-front and neutral. The results are compared to those obtained from averaging the output of the monodepth neural network model, which represents the state-of-the art. An overall increase of 14% in relative depth accuracy over relative depth computed from the monodepth model derived results is achieved.