Deep learning image data automatic annotation [easy to understand]

Hello everyone , I meet you again , I'm your friend, Quan Jun .

Tensorflow and Caffe Equal depth Study in , Supervised learning data Annotation is a very tedious and time-consuming work , At present, most companies use outsourcing to label companies for processing , Or buy an existing dataset , It makes the cost of in-depth learning and research extremely high . This paper introduces an idea and method of solving data annotation with artificial intelligence .

One 、 Ideas

step ：

1、 A preliminary model is used to detect the small batch of data to be labeled , The preliminary model here can be trained by yourself with a small batch of data sets , It can also be published online ;

2、 Correct the detected results by human intervention ;

3、 Train the new model with the corrected data ;

4、 Use the new model to detect the data to be tested in medium batch ;

5、 adopt 1~5 Loop iteration of steps , Can be refined step by step ;

6、 Although it also requires human participation , But it can greatly reduce the workload .

Implementation method ：

1、Anno-Mage

Anno-Mage It is a semi-automatic annotation tool , Test the data set through a general model . But the types of items that this tool can label are limited , There is no process of model iteration and gradual refinement , You can modify and optimize its source code by yourself .

github Code address ：https://github.com/virajmavani/semi-auto-image-annotation-tool

2、easyDL Intelligent annotation

2.1、 Intelligent annotation

Baidu easyDL It provides the function of intelligent annotation , Similar to the above ideas , All of them are labeled and trained for small batch data first , Then mark the remaining data sets with the learning results , Then correct it manually , Iterative refinement .

easyDL Website of the platform ：https://ai.baidu.com/easydl/lite

Intelligent detection technology documents ：https://ai.baidu.com/ai-doc/EASYDL/lk38n327g

2.2、 Export data

but easyDL The official does not provide data export function and api, This prevents us from getting the data Tensorflow and Caffe Training . So we can use crawler technology to crawl the trained data .

Tools github Address ：https://github.com/kooky126/easydl2labelImg

Publisher ： Full stack programmer stack length , Reprint please indicate the source ：https://javaforall.cn/147820.html Link to the original text ：https://javaforall.cn