Multi-data-set target detection training method and system, storage medium and electronic equipment

Abstract

The invention discloses a multi-data-set target detection training method and system, a storage medium and electronic equipment, and the method comprises the steps: respectively inputting a newly-added category data set and a corresponding first annotation file into a first detection model and a second detection model, and carrying out the detection model training; performing data cleaning on newly added categories in the historical data set by using the trained first detection model and second detection model to obtain a historical data set after data cleaning and a corresponding second annotation file; and merging the newly-added category data set and the historical data set after data cleaning, inputting the merged data set and a corresponding third annotation file into a third detection model, performing model training, and performing deployment after training is completed. According to the method, the manpower labeling cost, the model maintenance cost and the machine deployment cost are reduced, and the problem that the cost is relatively high when the newly added detection category in the historical data is not labeled in the prior art is solved.

Description

technical field [0001] The invention relates to the technical field of target detection, in particular to a multi-data set target detection training method, system, storage medium and electronic device. Background technique [0002] With the development of artificial intelligence technology, the application of computer vision technology in people's daily life is more and more extensive. As one of the most important tasks in computer vision, object detection plays a pivotal role. The general target detection defaults to limit the specific number of detected categories, but in actual needs, it is impossible to determine whether a new category is added. When a detection category is added, the historical data does not mark the newly added category to be detected. If the newly added category dataset is directly merged for model training, there are a large number of unlabeled new categories in the historical data set, which will cause the model to detect the historical category. ...

AI Technical Summary

Problems solved by technology

[0005] The main purpose of the present invention is to provide a multi-dataset target detection training method and system, using two detection models to clean the newly added categories in the historical data set, and merging multiple data sets into the third detection model for training to solve the problem of Existing technologies deal with the problem of high cost when new detection categories in historical data are not marked

Images