Target detection method and device, equipment and computer readable medium

Abstract

The invention discloses a target detection method and device and equipment and a computer readable medium. The method comprises the following steps: extracting an input feature image from an input image by utilizing a first convolutional neural network, wherein the size of the input feature image is smaller than that of the input image; performing convolution processing on the input feature imageby using a second convolutional neural network; and utilizing a third convolutional neural network to predict the convolutional input feature image output by the second convolutional neural network, and determining the position of the region where the target is located according to the prediction result output by the third convolutional neural network.

Description

technical field [0001] The present disclosure relates to the field of image processing, and in particular to a method, device, device and computer-readable medium for target detection performed by using a convolutional neural network. Background technique [0002] Methods based on deep learning are currently common target detection methods, such as two-stage methods such as RCNN, Fast-RCNN, Faster-RCNN or one-stage methods SSD, YOLO, DSSD, SSH, etc. Through the above-mentioned general object detection method, it is possible to detect multiple categories of objects, such as human faces. For example, by setting an "anchor" (anchor) used as a candidate area to regress the position of the target and classify the category of the target. [0003] In a usage scenario where only a single category of target detection is required, a detector designed for a specific target (such as a face detector MRCNN, S3FD, etc.) can be used for specific target detection. [0004] However, existin...

AI Technical Summary

Problems solved by technology

[0004] However, the existing object detection models all have the disadvantages of slow speed and large models.

Images