| Full text | |
| Author(s): |
Bulach Gapski, Marina Chagas
;
Valem, Lucas Pascotti
;
Guimaraes Pedronette, Daniel Carlos
Total Authors: 3
|
| Document type: | Journal article |
| Source: | 2024 37TH SIBGRAPI CONFERENCE ON GRAPHICS, PATTERNS AND IMAGES, SIBGRAPI 2024; v. N/A, p. 6-pg., 2024-01-01. |
| Abstract | |
In recent years, the volume of multimedia data has been rapidly increasing across various applications. Consequently, classification methods capable of handling scenarios with limited labeled data (e.g., semi-supervised, weakly supervised) have become critically important, particularly because acquiring labeled data is often expensive and time-consuming. Regarding image data, feature extraction approaches are commonly employed in many tasks. Feature extraction involves identifying and extracting key characteristics or patterns, such as edges, textures, shapes, and colors. Nowadays, most extractors consider deep learning strategies, such as Convolutional Neural Networks (CNNs) and Vision Transformers (ViT). With various feature extractors available in the literature, there is a wide diversity of features that can be considered. The features extracted from an image depend on the application, the extractor used, and its configuration. Therefore, combining different extractors can be a promising strategy to exploit complementary information. Graph Convolutional Networks (GCNs) are fundamental and promising strategies in the scenario of semi-supervised image classification, being able to leverage labeled and unlabeled data, and exploiting the graph structures that offer valuable information. This work proposes an approach for GCNs in scenarios where labeled data is scarce, combining sets of features and graphs considering different extraction approaches. Among the main contributions, the experimental results reveal that these combinations and the use of manifold learning to process these graphs improve the classification results in most cases. (AU) | |
| FAPESP's process: | 23/12736-3 - Investigation of Visual Question Answering Approaches for Natural Disaster Management |
| Grantee: | Marina Chagas Bulach Gapski |
| Support Opportunities: | Scholarships in Brazil - Master |
| FAPESP's process: | 18/15597-6 - Aplication and investigation of unsupervised learning methods in retrieval and classification tasks |
| Grantee: | Daniel Carlos Guimarães Pedronette |
| Support Opportunities: | Research Grants - Young Investigators Grants - Phase 2 |