Iván García, Jorge García González, Rafael Marcos, Ezequiel López Rubio, Enrique Domínguez Merino
Image segmentation in computer vision applications plays a critical role in the video processing workflow. In real applications, where interesting elements are moving in the presence of moving objects in the background, complex models are required in the segmentation process to obtain better results. In this paper, a methodology based on super-resolution and test time augmentation is proposed to improve the precision and effectiveness of the segmentation process. Our proposal avoids both modification and retraining of the model. Experiments show that our approach can increase the mean average precision of images segmentation in sequences from well-known benchmark datasets with a significant improvement.
© 2001-2024 Fundación Dialnet · Todos los derechos reservados