Scene segmentation network training method, device, computing equipment and storage medium

Abstract

The invention discloses a scene segmentation network training method, device, computing equipment and computer storage medium, wherein the method is completed through multiple iterations; sample images are extracted and scene segmentation results are marked; the sample images are input into the scene segmentation network to perform Training, wherein, in the scene segmentation network, at least one layer of convolutional layer is used to scale the first convolutional block of the convolutional layer by using the scale coefficient output by the scale regression layer to obtain the second convolutional block, and then use the second The convolution block performs the convolution operation of the convolutional layer to obtain the output result of the convolutional layer; obtains the corresponding sample scene segmentation result; updates the scene segmentation network according to the segmentation loss between the sample scene segmentation result and the labeled scene segmentation result The weight parameter of ; execute the above training steps iteratively until the predetermined convergence condition is met. The technical solution realizes the adaptive scaling of the receptive field, and improves the accuracy and processing efficiency of image scene segmentation.

Description

technical field [0001] The present invention relates to the technical field of image processing, in particular to a scene segmentation network training method, device, computing equipment and computer storage medium. Background technique [0002] In the existing technology, the training of the segmentation network is mainly based on the fully convolutional neural network in deep learning, using the idea of ​​transfer learning to migrate the network pre-trained on the large-scale classification data set to the image segmentation data set Training is performed to obtain a segmentation network for scene segmentation. [0003] In the prior art, the network architecture used when training the segmentation network directly uses the image classification network. The size of the convolutional block in the convolutional layer is fixed, so the size of the receptive field is fixed. Among them, the receptive The field refers to the area of ​​the input image corresponding to the respons...

AI Technical Summary

Problems solved by technology

However, for image scene segmentation, the scene often contains objects of different sizes. Using a segmentation network with a fixed-size receptive field often causes problems when dealing with too large and too small objects. For example, for smaller objects, the receptive field It will capture too much background around the target, thereby confusing the target with the background, causing the target to be missed and misjudged as the background; for larger targets, the receptive field can only capture a part of the target, which makes the target category judgment biased, resulting in Discontinuous Segmentation Results

Therefore, the segmentation network trained in the prior art has the problem of low accuracy of image scene segmentation

Images