Target detection method and apparatus, training method, electronic device and medium

A target detecting network training method, comprising:

• separately extracting, by means of a neural network, characteristics of a template frame and a detection frame, wherein the template frame is a detection box image of a target object, and the template frame is smaller than the detection frame in image size;
• performing a convolution operation on the characteristic of the template frame by means of a first convolutional layer, wherein a first characteristic obtained by the convolution operation is used as a classification weight of a local region detector;
• and performing the convolution operation on the characteristic of the template frame by means of a second convolutional layer, wherein a second characteristic obtained by the convolution operation is used as a regression weight of the local region detector;
• inputting the characteristic of the detection frame into the local region detector to obtain classification results and regression results of multiple alternative boxes output by the local region detector;
• obtaining a detection box for the target object in the detection frame according to the classification results and regression results of the multiple alternative boxes output by the local region detector;
• and
• taking the obtained detection box for the target object in the detection frame as a predictive detection box, training the neural network, the first convolutional layer, and the second convolutional layer based on label information of the detection frame and the predictive detection box.

An electronic device, comprising:

• a memory, configured to store executable instructions;
• and
• a processor, configured to communicate with the memory to execute the executable instructions so as to:
• separately extract, by means of a neural network, characteristics of a template frame and a detection frame wherein the template frame is a detection box image of a target object, and the template frame is smaller than the detection frame in image size;
• obtain a classification weight and a regression weight of a local region detector based on the characteristic of the template frame;
• input the characteristic of the detection frame into the local region detector to obtain classification results and regression results of multiple alternative boxes output by the local region detector;
• and
• obtain a detection box for the target object in the detection frame according to the classification results and regression results of the multiple alternative boxes output by the local region detector.