| spelling |
Douglas Guimarães Macharethttp://lattes.cnpq.br/7640548709008824Erickson Rangel do NascimentoWilliam Robson SchwartzSandra Eliza Fontes de Ávilahttp://lattes.cnpq.br/0036531494047684Eduardo Vieira e Sousa2022-09-13T15:00:04Z2022-09-13T15:00:04Z2021-07-29http://hdl.handle.net/1843/45142Modern societies are composed of complex structures that emerge from the relationships between individuals, and the comprehension of these arrangements has the potential to become a powerful tool for intelligent systems. Current image-based social relation recognition methods isolate specific information from the input to capture essential aspects defining these relationships. However, this is an inaccurate approach for analyzing social relations, since the interaction between all these parts form an intricate structure, which is as valuable as the information each piece carries individually. For this reason, it is crucial to capture the original social structure to achieve the high-level reasoning required to identify relationships adequately. In this work, a novel approach to interpret image-based social relation recognition methods is presented, considering three distinct scopes named social scales, regarding individual, relative, and general information. Additionally, it also evaluates how the data from these different social perspectives is combined, taking into account the capacity of capturing multi-scale interdependencies and constraints. The Social Knowledge Graph (SKG) is proposed based on the conclusions obtained from the conducted analysis, producing a representation capable of replicating the original social structure from the input image. This unique representation is exploited with the Social Graph Network (SGN) by employing specific feature aggregation strategies according to the information embedded in the graph structure. The performance of the proposed method was evaluated in well-known benchmarks for social relation recognition, achieving a new state-of-the-art. Finally, a deep analysis of the methodology and the main concepts behind it is conducted, providing insight into the decision-making process of the proposed model and delivering results that support the new interpretation of social relationship recognition methods.As sociedades modernas são compostas por estruturas complexas que emergem das relações entre indivíduos, e a compreensão desses arranjos tem o potencial de se tornar uma ferramenta poderosa para sistemas inteligentes. Os métodos atuais de reconhecimento de relações sociais baseados em imagens isolam informações específicas da entrada com intuito de capturar aspectos essenciais que definem esses relacionamentos. No entanto, esta é uma abordagem imprecisa para analisar relações sociais, uma vez que a interação entre todas essas partes forma uma estrutura intrincada, sendo tão valiosa quanto as informações que cada componente carrega individualmente. Por esse motivo, é crucial capturar a estrutura social original para alcançar o raciocínio de alto nível necessário para identificar os relacionamentos de forma adequada. Neste trabalho, uma nova abordagem para interpretar métodos de reconhecimento de relacionamento social baseados em imagens é apresentada, considerando três escopos distintos de análise, denominados escalas sociais, relacionados a informações individuais, relativas e gerais. Além disso, também é avaliado como os dados dessas diferentes perspectivas sociais são combinados, levando em conta a capacidade de capturar dependências e restrições em múltiplas escalas. O Social Knowledge Graph (SKG) é proposto com base nas conclusões obtidas da análise conduzida, produzindo uma representação capaz de replicar a estrutura social da imagem de entrada. Essa representação única é explorada por meio da Social Graph Network (SGN), aplicando estratégias específicas de agregação de features, conforme as informações embutidas na estrutura do grafo. O desempenho do método proposto foi avaliado em benchmarks bem estabelecidos, alcançando um novo estado da arte. Finalmente, uma análise profunda da metodologia e dos conceitos por trás dela é conduzida, fornecendo uma visão sobre o processo de decisão do modelo proposto e obtendo resultados que suportam a nova interpretação dos métodos de reconhecimento de relações sociais.CAPES - Coordenação de Aperfeiçoamento de Pessoal de Nível SuperiorengUniversidade Federal de Minas GeraisPrograma de Pós-Graduação em Ciência da ComputaçãoUFMGBrasilICX - DEPARTAMENTO DE CIÊNCIA DA COMPUTAÇÃOComputação – TesesVisão computacional – TesesReconhecimento de padrões – TesesAprendizado profundo – TesesRedes de relações sociais – TesesRedes neurais de grafos – TesesComputer visionPattern recognitionSocial relation recognitionDeep learningGraph neural networksStructure-aware multi-scale reasoning for image-based social relation recognitionRaciocínio estrutural multiescala para reconhecimento de relações sociais em imagensinfo:eu-repo/semantics/publishedVersioninfo:eu-repo/semantics/masterThesisinfo:eu-repo/semantics/openAccessreponame:Repositório Institucional da UFMGinstname:Universidade Federal de Minas Gerais (UFMG)instacron:UFMGORIGINALStructure-aware multi-scale reasoning for image-based social relation recognition.pdfStructure-aware multi-scale reasoning for image-based social relation recognition.pdfapplication/pdf74510901https://repositorio.ufmg.br/bitstream/1843/45142/3/Structure-aware%20multi-scale%20reasoning%20for%20image-based%20social%20relation%20recognition.pdf773d068dc8c98979dba3e1330d643c7bMD53LICENSElicense.txtlicense.txttext/plain; charset=utf-82118https://repositorio.ufmg.br/bitstream/1843/45142/4/license.txtcda590c95a0b51b4d15f60c9642ca272MD541843/451422022-09-13 12:00:04.635oai:repositorio.ufmg.br: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ório InstitucionalPUBhttps://repositorio.ufmg.br/oaiopendoar:2022-09-13T15:00:04Repositório Institucional da UFMG - Universidade Federal de Minas Gerais (UFMG)false
|