一种基于Labelimg的辅助标注方法

随着目标识别技术的发展,YOLO神经网络被应用在各个领域中,但是YOLO自带的预训练模型只能广泛地识别目标,具有普遍性,缺少特殊性。因此,需要自行制作具有针对性的样本训练集。对于YOLO的样本训练集,在样本图像数量上包含了几千甚至数万张图片,其中包含的对象类别也数以百计。在公开的数据集中难以找到符合该文要求的数据集,这就需要人工制作样本数据集。在传统的人工标注方法中,常使用的标记工具有labelimg、labelme等标注工具,但是传统的标注方法中存在着对象标记工作量大、标注效果精度低等缺点,该文引入SAM模型辅助标注方法,这种标注方法在能够高速完成大量标注工作的同时,标注的效果精度也十分理想。

With the development of target recognition technology,YOLO neural network is used in various fields,but YOLO comes with a pre-training model can only widely recognize the target,with a universal lack of specificity.Therefore,it is necessary to create a targeted sample training set for YOLO's sample training set,the number of sample images contains thousands or even tens of thousands of images,which contains hundreds of object categories.It is difficult to find a data set that meets the requirements of this paper in the public data set,which requires manual production of sample data sets,in the traditional manual labeling methods,often used labeling tools are labelimg,labelme and other labeling tools,but the traditional labeling methods exist in the object labeling workload,the labeling effect of the low-precision drawbacks such as low accuracy,this chapter introduces a SAM model-assisted labeling method,which can complete a large number of objects at high speed and with high accuracy.This labeling method can complete a large number of labeling work at high speed,and the precision of the labeling effect is also very ideal.