Real-time instance segmentation method and system based on single-stage network and electronic equipment

Abstract

The invention discloses a real-time instance segmentation method and system based on a single-stage network and electronic equipment. The real-time instance segmentation method based on the single-stage network comprises the following steps: performing target detection processing on an original image through a single-stage target detection network model to obtain a series of feature maps and detection target data; pre-processing a part of the feature maps in the series of feature maps to obtain corresponding reference feature maps; taking a frame of a detection target in the detection target data as a region of interest, performing cutting and splicing processing on the reference feature map to obtain a spliced feature map; and performing mask prediction processing on the spliced feature map through a mask prediction network model to obtain a mask segmentation result corresponding to the detection target.

Description

technical field [0001] The invention relates to the technical field of image segmentation, in particular to a single-stage network-based real-time instance segmentation method, system and electronic equipment. Background technique [0002] Currently, image segmentation tasks in computer vision usually include semantic segmentation and instance segmentation. The semantic segmentation generally refers to classifying pixels according to the semantic meaning expressed in the image, so that the computer can identify the target category of each pixel and mark the corresponding label, so as to realize the understanding of the visual scene. The instance segmentation is based on semantic segmentation. In addition to identifying the target category of each pixel, it also needs to separate different instances of the same category. [0003] With the great success of Convolution Neural Network (CNN in English) in visual understanding, in various applications such as wearable devices (su...

AI Technical Summary

Problems solved by technology

However, due to the large number of candidate regions and the complex calculation of the screening process, the existence of the region recommendation stage will greatly increase the calculation time of the existing mask region recommendation convolutional neural network, making the existing mask region It is recommended that convolutional neural networks are difficult to achieve real-time instance segmentation, and cannot meet the increasingly high real-time requirements in the fields of augmented reality, intelligent robots, and unmanned driving.

Images