Instance segmentation method and device based on feature attention and sub-upsampling

Abstract

The invention discloses an instance segmentation method based on feature attention and sub-upsampling. The method comprises the steps of obtaining a to-be-segmented original image; extracting a feature map from the original image through a feature global network, and determining a region of interest in the feature map, the feature global network comprising an attention module; aligning and extracting a region of interest from the feature map; and classifying the extracted regions of interest, and generating segmentation masks for the extracted regions of interest by utilizing sub-pixel up-sampling so as to realize instance segmentation of the original image. The method has the advantages that the attention module is added when the feature map is extracted; according to the embodiment of the invention, redundant information and fusion information can be deleted by applying channel transformation after maximum pooling and average pooling operations, the effectiveness of an image featurevector is improved. Meanwhile, the accuracy of segmentation and detection in instance segmentation is improved under the condition of not reducing the speed in combination with a sub-pixel up-samplingmode, and the occupied memory is not increased.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to an instance segmentation method and device based on feature attention and sub-upsampling. Background technique [0002] Instance segmentation is a kind of pixel-by-pixel segmentation, which is a further development of semantic segmentation. Semantic segmentation does not distinguish between different instances belonging to the same category. For example, when there are multiple cats in an image, semantic segmentation predicts all pixels of two cats as "cat". And instance segmentation further needs to distinguish which pixels belong to the first cat and which pixels belong to the second cat. [0003] Faster-RCNN is an earlier target detection network. For an input picture, the network can get a list of bounding boxes, the category label of each bounding box, and the probability of each bounding box category label. It contains a convolutional layer for extra...

AI Technical Summary

Problems solved by technology

[0008] An embodiment of the present invention provides an instance segmentation method and device based on feature attention and sub-upsampling, which is used to solve the problems of low effectiveness of image feature vectors, complex calculations and low accuracy in instance segmentation methods in the prior art. The method includes:

Images