Weak supervision detection model training method and system based on collaborative learning

Abstract

The invention discloses a weak supervision detection model training method and system based on collaborative learning. The weak supervision detection model training method comprises the steps of: extracting abstract visual features of an image by using a convolutional neural network; then carrying out prediction on a detection result according to visual features by using two types of strong and weak detection models, wherein a strong detector uses bounding box labels as training conditions, and a weak detector only uses an image-level label as a training condition; then calculating a trainingerror, wherein the error consists of two parts of a strong detector error and a weak detector error, the strong detector error is defined by consistency loss, and the weak detector error is image-level cross entropy loss; and then simultaneously updating parameters of the strong and weak detectors by utilizing the training error, until training is converged. According to the invention, two detection models with different properties are simultaneously trained on the basis of a collaborative learning mode, and two detection models collaborate with each other and are improved together in the training process, so that a problem that the detection model is difficult to train under the weak supervision condition is solved.

Description

technical field [0001] The present invention relates to the field of computer vision, in particular to a method and system for training a weakly supervised detection model based on collaborative learning. Background technique [0002] Computer vision, especially target detection technology, plays a vital role in specific fields, such as surveillance capture, unmanned vehicles, etc. With the help of deep learning, a large number of successful detection models have emerged in recent years, and these detection models far exceed the previous detection models in terms of recognition accuracy and speed. However, training a high-accuracy detection model requires a large amount of precisely labeled image data as model supervision conditions, which often requires a lot of manpower and material resources. At the same time, a large amount of roughly labeled data can be easily obtained from the Internet, so a very practical research direction is to study how to effectively train the de...

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to overcome the deficiencies of the prior art and provide a weakly supervised detection model training method and system based on collaborative learning to solve the problem that the detection model is difficult to train under the condition of only providing rough picture category labels

Images