Publicação
Transformers in Object Detection: A comparative analysis of the effectiveness of transformer-based neural networks in Object Detection tasks
| datacite.subject.fos | Ciências Naturais::Ciências da Computação e da Informação | pt_PT |
| dc.contributor.advisor | Castelli, Mauro | |
| dc.contributor.advisor | Costa, Victor Cardoso Reis | |
| dc.contributor.author | Neves, Matias Marques Condessa de Sousa | |
| dc.date.accessioned | 2024-02-15T16:35:44Z | |
| dc.date.available | 2025-01-29T01:33:16Z | |
| dc.date.issued | 2024-01-29 | |
| dc.description | Dissertation presented as the partial requirement for obtaining a Master's degree in Data Science and Advanced Analytics, specialization in Data Science | pt_PT |
| dc.description.abstract | Artificial Intelligence is applicable to many different tasks, one of those tasks being Object Detection. For this task, the most common Machine Learning models are Artificial Neural Networks. While many Artificial Neural Networks are experimented with, Convolutional Neural Networks are currently the most common and most developed models that can be used in Object Detection. However, Artificial Intelligence models that interpret data using different sets of algorithms are still consistently researched, namely Transformer-based models. In this thesis, Artificial Intelligence models that integrate Transformer architectures are compared with a state-of-the-art Convolutional Neural Network, the comparisons being based on three Object Detection tasks: Identifying blueberry batches in images, identifying balloons in images, and identifying objects represented in the COCO2017 dataset. The first task is the most extensively researched one: it’s a real-world Precision Agriculture task, and four datasets were annotated from existing footage to train the Object Detection models to perform this task. This task involves a great deal of experimental work, so the other two tasks serve as more stable benchmarks: the second task involves an open-source dataset, and for the third task comparisons were made between already existing results of training these models. Models are compared based on their predictive power and their training time. This thesis aims to determine whether Transformer-based Artificial Intelligence models can feasibly compete with state-of-the-art solution in Object Detection tasks, and, consequently, whether further research on the use of Transformers for Object Detection can be expected to be fruitful in a practical sense. | pt_PT |
| dc.identifier.tid | 203518373 | pt_PT |
| dc.identifier.uri | http://hdl.handle.net/10362/163583 | |
| dc.language.iso | eng | pt_PT |
| dc.subject | Transformers | pt_PT |
| dc.subject | Object Detection | pt_PT |
| dc.subject | Neural Networks | pt_PT |
| dc.subject | Artificial Intelligence | pt_PT |
| dc.subject | Precision Agriculture | pt_PT |
| dc.subject | SDG 2 - Zero hunger | pt_PT |
| dc.title | Transformers in Object Detection: A comparative analysis of the effectiveness of transformer-based neural networks in Object Detection tasks | pt_PT |
| dc.type | master thesis | |
| dspace.entity.type | Publication | |
| rcaap.rights | openAccess | pt_PT |
| rcaap.type | masterThesis | pt_PT |
| thesis.degree.name | Mestrado em Ciência de Dados e Métodos Analíticos Avançados, especialização em Ciência de Dados | pt_PT |