Model training method and device, target detection method and device, equipment and storage medium

Abstract

The embodiment of the invention discloses a model training method and device, a target detection method and device, equipment and a storage medium. The method comprises the steps of acquiring a first sample with an instance-level label and a second sample with an image-level label; the second sample is obtained based on the instance-level label of the first sample; determining a pseudo label of sample data in the second sample through a pre-trained target detection model; determining the original detection loss of the target detection model based on the instance-level label of the sample data in the first sample; determining classification enhancement loss of the target detection model based on a pseudo label of sample data in the second sample; and training the target detection model by using the first sample and the second sample based on the original detection loss and the classification enhancement loss.

Description

technical field [0001] This application relates to the field of computer vision, involving but not limited to model training methods, object detection methods, devices, equipment and storage media. Background technique [0002] At present, the mainstream target detection technology is applied in the scene where the data is relatively balanced, but in reality, as the number of categories to be detected increases, the categories will naturally show a long-tail distribution. This serious class imbalance problem will greatly affect the performance of object detectors. [0003] For long-tail target detection, the current mainstream method is to use loss reweighting and data resampling, but these methods are still performed in the case of limited data sets. For pictures with some minority categories, the amount of data may be very small, which makes the neural network, which is very data-dependent, unable to show good generalization performance on these minority categories. Con...

AI Technical Summary

Problems solved by technology

This serious class imbalance problem will greatly affect the performance of the target detector

[0003] For long-tail target detection, the current mainstream method is to use loss reweighting and data resampling, but these methods are still carried out in the case of limited data sets

For pictures with some minority categories, the amount of data may be very small, which causes the neural network, which is very data-dependent, to fail to show good generalization performance on these minority categories.

Images