Target segmentation system and its training method, target segmentation method and equipment

Abstract

The invention discloses a target segmentation system, its training method, target segmentation method and equipment. The system includes: a semantic perception network, which adopts the form of a fully convolutional network, including a convolution module, a pooling module and a regularization module, and the semantic perception network is set to extract the semantic feature map of the image; a texture perception network, which adopts a network without pooling Form, including serially arranged dilated convolutional layer, feature shrinkage layer, feature expansion layer and the first convolutional layer, the texture perception network is set to extract the texture feature map of the image; the feature fusion layer is set to extract the semantic feature map Splicing and merging with the texture feature map to obtain the target segmentation map of the image. The invention proposes a dual-branch multi-scale feature fusion model, which improves the accuracy and robustness of multi-scale object segmentation in natural images.

Description

technical field [0001] Embodiments of the present invention relate to the field of computer vision, and in particular to an object segmentation system and its training method, an object segmentation method and equipment. Background technique [0002] Image segmentation is a classic problem in the field of computer vision, and it is one of the important ways to complete scene understanding. More and more applications and scenarios acquire knowledge from images, such as autonomous driving, human-computer interaction, intelligent robots, and augmented reality, etc., which highlights the importance of image segmentation as a core problem in computer vision. Image segmentation can be defined as a specific image processing technique used to divide an image into two or more meaningful regions. Image segmentation can also be seen as the process of defining boundaries between various semantic entities in an image. From a technical perspective, image segmentation is the process of a...

AI Technical Summary

Problems solved by technology

However, the depth of the network pursued in the current deep learning will make this structure ignore the shallow texture information, which will inevitably cause the network to lack the ability to perceive multi-size objects.

Images