Efficient full integer quantization method for image detection model

Abstract

The invention discloses an efficient full-integer quantization method for a target detection model. The weight, bias, input feature map and output feature map of each convolution layer in an image detection model are all represented through integers, wherein the integer calculation is carried out in a quantization reasoning process. The method specifically comprises the following steps: carrying out normal training, quantitative perception training and quantitative parameter generation on an image detection model of a real number version, and carrying out reasoning based on full integer operation on computing equipment by applying generated parameters of each layer. The method can greatly reduce the reasoning time of the image detection model, reduces the space of the model in the aspectsof disk storage and memory occupation, maintains the high detection precision of the image detection model, and facilitates the implementation of a more efficient image target detection system on FPGAand other computing devices.

Description

technical field [0001] The invention relates to a high-efficiency full-integer quantization method for an image detection model, which belongs to the technical field of neural networks. Background technique [0002] In recent years, the convolutional neural network has made great progress in many visual tasks such as image target recognition, target detection, and pixel segmentation. It has achieved performance beyond human performance in many tasks, and many well-known network structures have emerged. Such as AlexNet, VGGNet, GoogleNet, ResNet, DenseNet, SENet, etc. The development of these convolutional networks is characterized by an increasing number of network layers and increasing capacity, resulting in an increase in network parameters and calculations, which is not conducive to deployment and application on small mobile devices with limited resources. For example, hardware platforms such as smartphones and drones hope that the model should be as small as possible wh...

AI Technical Summary

Problems solved by technology

However, it does not take full advantage of the symmetry in the case of symmetric quantization, making the calculation process cumbersome

Images