An image instance segmentation method, device, electronic device and storage medium

Abstract

The present disclosure relates to an image instance segmentation method, device, electronic device and storage medium, and relates to the field of computer vision technology to solve the problem of inaccurate instance segmentation in the related art. The disclosed method includes: inputting a feature map of a target into a training In the segmentation network, the sub-image of the target in the feature map is determined based on the segmentation network, and the segmentation network is trained as follows: the sample image is down-sampled, and the down-sampled sample image is processed based on the hole convolution kernel of different expansion rates. Process, determine the sample maps of different receptive fields, fuse the sample maps of different receptive fields, obtain the fused sample map, perform up-sampling and instance segmentation on the fused sample map, determine the first sub-image, according to the first sub-image. One sub-image and the labeled second sub-image train the initial segmentation network; the position of the sub-image in the image is labeled, because the pixel information of the pixel points in the feature maps of different receptive fields is more, so the instance segmentation is more accurate.

Description

technical field

[0001] The present disclosure relates to the technical field of computer vision, and in particular, to an image instance segmentation method, apparatus, electronic device, and storage medium. Background technique

[0002] Instance segmentation is an important part of computer vision, and many research work has also been carried out to improve the segmentation accuracy. The earliest instance segmentation method is a bottom-up method, such as the Deepmask method, which first predicts the candidate region for segmentation, and then uses the classification network for classification. However, in this method, the segmentation process precedes the classification process, resulting in a slow instance segmentation process. , the accuracy is not high. The FCIS (Fully Convolutional Instance-awareSemantic Segmentation, fully convolutional instance-aware semantics) scheme that appeared next combines the segmentation candidate region and the object detection system, whic...

Images