Point cloud-oriented efficient binary neural network quantization method and device

Abstract

The invention discloses a point cloud-oriented efficient binary neural network quantization method and a device. According to the method, an entropy-maximized aggregation function and a layer-by-layer scale recovery step are used, so that the point cloud neural network can achieve information entropy maximization, network weight and activation quantization loss minimization by minimizing a loss function in a training process, thereby reducing quantization loss and improving the performance of the binary point cloud neural network. The method is completely compatible with bitwise operation, and has the advantage of quickly deducing and quantifying the neural network. The fact shows that the method has excellent performance in various network architectures, is superior to the prior art in the aspects of memory consumption, reasoning speed and accuracy, and is particularly suitable for realizing target detection in an automatic driving scene so as to meet the real-time detection positioning requirement of an automatic driving vehicle.

Description

technical field [0001] The present invention relates to a high-efficiency binarization neural network quantification method for point clouds, and also relates to a target detection device that adopts the neural network quantification method for the needs of automatic driving scenes, and belongs to the field of deep learning technology. Background technique [0002] Point cloud, especially point cloud neural network, has received more and more attention in various computer vision applications, such as autonomous driving, augmented reality, etc. Traditional point clouds usually have massive parameters and high computational completeness, and the training and inference process for a single task takes a lot of time. The main reason for this problem is that the models that currently achieve the best results on various tasks generally use neural networks with full precision, which makes these models need to use a large amount of storage resources. At the same time, many applicati...

AI Technical Summary

Problems solved by technology

This largely limits the use of point clouds on portable devices

Especially in autonomous driving scenarios, the computing power and storage capacity of on-board computing devices are often limited, which seriously restricts the application prospects of point cloud neural networks in this scenario.

Images