Interactive method and system for semi-automatic image annotation

Abstract

The invention relates to an interactive method of image semi-automatic annotation, comprising S1 dividing an initial sample into three different types of annotation samples according to different category attributes; labeling the three types of labeling samples manually to get different kinds of labeling results, and then using three models of Mask-RCNN, Fast-RCNN and FCN to train separately; S2 processing the data set of the picture to be annotated in an offline manner, wherein the annotating process is that the data set of the picture to be annotated passes through the three depth learning models in turn to output the json format files of all types and coordinate points of the data samples; S3 calling the relevant attribute tag value and coordinate point value of the json format file according to the name of the annotated image; S4 displaying the corresponding automatic marking result in the marking software, and judging whether the category and area marking of the target object arestandardized and reasonable by manpower; S5 carrying out data augmentation on the correctly labeled labeling samples and feeding back the augmented data to the model for retraining.

Description

technical field [0001] The invention relates to the technical field of auto-driving road image tagging, in particular to an interactive method and system for semi-automatic tagging of images. Background technique [0002] Data is the fuel of AI, which fully demonstrates the importance of data in the field of autonomous driving is self-evident, especially since the domestic autonomous driving started late and lacks reasonable and effective data sets for training. In complex road conditions, autonomous driving is far from meeting the road standards, such as: the identification and positioning of vehicles, pedestrians, and road signal lights in different road environments. Many of these problems are difficult to solve only by technical means. Therefore, the help of large-scale and accurate data sets is needed. The existing methods for labeling data can be divided into two types: traditional manual labeling and semi-automatic intelligent labeling. [0003] Traditional manual la...

AI Technical Summary

Problems solved by technology

However, the actual data to be labeled is a very difficult process. The labeling task is complex, cumbersome, and time-consuming, and it takes a lot of time to organize and label the data; There will also be some differences in cognition, and some labeling errors that do not meet the standards may occur during the labeling process, and these non-standard parts of the labeling will bring huge challenges to the subsequent deep learning tasks

[0004] Semi-automatic intelligent labeling is the application of deep learning technology to automatic labeling technology. Recently, deep learning has set off a new wave of artificial intelligence. It is widely used in various fields such as unmanned driving, medicine, face recognition and speech understanding and translation. The labeling process is to train a large number of standardized road datasets through the deep learning neural network, and use the obtained deep learning model to identify and locate unlabeled road target features. However, in the case of insufficient training datasets and large numbers of dataset categories Under certain circumstances, the effect of partial automatic labeling may be different from the actual coordinate point position of the ground object. This is mainly due to the insufficient generalization ability of a single deep learning model in the process of training the model, and it may also be due to underfitting or Over-fitting, so it is necessary to manually review and correct the correct attributes and coordinate point positions of the ground objects. Although this method improves the labeling efficiency to a certain extent, it is still not suitable for sample labeling under large samples and multiple labels.

Images