Instance segmentation method based on period B spline

Abstract

The invention discloses an instance segmentation method based on a period B spline. Reversely solving a period B spline control point by acquiring an object contour coordinate point of the picture; length representation and angle representation of the periodic B-spline control points are obtained by combining the periodic B-spline control points with neural network regression; establishing a Gaussian heat map, a loss function and a target construction formula for neural network training; cartesian coordinates of a periodic B spline control point are obtained, the vectorization purpose is achieved by conducting periodic B spline modeling on an object contour, periodic B spline control point information is returned through a neural network, and therefore vectorization representation of the object contour can be quickly and accurately obtained under the condition that manual intervention is not needed.

Description

technical field [0001] The invention relates to an instance segmentation technique, in particular to an instance segmentation method based on periodic B-splines. Background technique [0002] Instance segmentation is one of the basic tasks of computer vision. It can not only accurately classify objects, but also needs to give the location mask of objects. In recent years, with the application of deep learning in computer vision, based on convolutional neural network The accuracy of the instance segmentation method on related data sets is getting higher and higher, but at the same time, the model structure is becoming more and more complex, and the speed and memory usage cannot meet the actual application requirements. The existing instance segmentation method It is mainly divided into three categories: detection-based, segmentation-based and contour-based. The detection-based instance segmentation method first uses the detector to detect the bounding box of the object, and t...

AI Technical Summary

Problems solved by technology

[0003] In order to overcome the deficiencies of the prior art, the present invention provides an instance segmentation method based on periodic B-splines that obtains the vectorized representation of the object contour without increasing the complexity of the model

Images