Compressive attention model for semantic segmentation of pixel group

Abstract

The invention discloses a compressed attention model for semantic segmentation of pixel groups. The model utilizes an effective compressed attention (SA) module to solve two unique features of the pixel groups in semantic segmentation: (1) pixel group attention and (2) pixel-by-pixel prediction. Specifically, according to the provided SA model, attention of a pixel group is added to conventional convolution by introducing an attention convolution channel, so that interdependence of space channels is considered in an effective mode. Unlike existing attention models, a compressed global attention model (SA) is generated while aggregating multi-scale features using downsampling channels implemented through pooling layers. Therefore, the SA model enhances the target of pixel-level dense prediction and takes into account the problem of ignored pixel group attention. Compared with other methods, the invention has the advantages that the method is obviously improved, and the test result on a PASCAL data set shows that the PACc and mIoU of the new method are higher than those of classical methods such as FCN50 and FCN101.

Description

technical field [0001] The invention belongs to the technical field of semantic segmentation, and in particular relates to a compressed attention model for semantic segmentation pixel groups. Background technique [0002] Convolutional Network (FCN) recovers the category to which each pixel belongs from abstract features. That is, the classification from the image level is further extended to the classification at the pixel level. The deconvolution layer is used to upsample the feature map of the last convolutional layer to restore it to the same size as the input image, thus producing a prediction for each pixel while preserving the spatial information in the original input image , and finally perform pixel-wise classification on the upsampled feature map. However, most of them focus on improving segmentation performance from the pixel level, but largely ignore the implicit task of pixel grouping. [0003] Multi-Scale Spatial Asymmetric Recalibration (MS-SAR), which demo...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the deficiencies in the prior art, the technical problem solved by the present invention is to provide a kind of compressed attention model for semantically segmented pixel groups, and to solve the problem of considering neglected pixel groups attention

Images