Target detection model training method based on knowledge distillation

Abstract

The invention provides a target detection model training method based on knowledge distillation, and the method comprises the steps: training a target detection teacher model through a training sample image set, and a training sample image comprises a first label which is a target detection frame center point pixel position hard label probability matrix; a second label which is the width and the height of the target detection frame; and a third label which is the offset of the pixel position of the central point of the target detection frame. The prediction output result of the target detection teacher model comprises a target detection frame center point pixel position probability thermodynamic diagram, the width and height of the target detection frame, and the target detection frame center point pixel position offset; after the loss function of the target detection student model is improved in a knowledge distillation mode, the target detection student model is trained and generated. According to the method, the problems that the target detection model obtained by training through an existing knowledge distillation method cannot simultaneously ensure that the network structure is simple and the use requirement of terminal equipment cannot be met, and the recognition rate of the target detection model is excellent to ensure the model detection precision are solved.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence model training, in particular to a knowledge distillation-based target detection model training method. Background technique [0002] Knowledge distillation is to guide the training of the network structure of the student model by introducing the network structure of the teacher model, and then realize knowledge transfer. The specific method steps are to train the teacher model first, and then use the output of the teacher model and the real labels of the data to train the student model, so as to transfer the knowledge of the network structure of the teacher model to the network structure of the student model, ensuring that the student model The network structure can obtain the performance close to the network structure of the teacher model, and also make the network structure of the student model as small as possible, with fewer parameters, which is more conducive to reducing the ...

AI Technical Summary

Problems solved by technology

[0006] The main purpose of the present invention is to provide a target detection model training method based on knowledge distillation, so as to solve the problem that the target detection model obtained by training with the knowledge distillation method in the prior art cannot simultaneously ensure that the network structure is simple and meet the needs of terminal equipment, and The recognition rate of the target detection model is excellent to ensure the accuracy of model detection

Images