engleski

Convolutional Models for Segmentation and Localization

The revival of deep models has profoundly improved the accuracy of image classification models and provided a large improvement potential in related computer vision tasks. Recently, much attention has been directed to­ wards dense prediction models which produce distinct output in each image pixel. This paper addresses two particular instances of dense prediction: object loca­lization and semantic segmentation. We briefly review the underlying operation principles, present some of our experimental results and discuss ways to analyze the success of learning and the utility of the resulting models.

Convolutional models

nije evidentirano