AI learning is processed through training of examples and myriads of labeled data are needed in the process. Creating a simple AI model to classify images requires a thousand to tens thousand images. In many cases in the past, large amounts of data have been manually labeled, requiring a lot of time and effort. Some compared the process to 'sticking dolls' eyeballs.'There is another problem of labeling manually. Since tens of thousands of data cannot be manually labeled by a single person, the work is often divided among several people. And here, the consistency of labeling becomes a problem. As an example, let's say you need to create a model that classifies images of several animals. Several people will need to manually classify each image and label them with different animal classes. And then they see the following image.
Looking at this ambiguous picture, some would say it is a rabbit and other would say duck. If a single person were to label all of the images, the images would be labeled with consistently. However, the labeling process may lose consistency under several different people who all have different views and perspectives. But auto-labeling could be a solution to all these problems (time, man power, and consistency).
Auto-labeling is processed by taking a very small amount of labeled data and automatically labeling the rest of the data. Let's see how Click AI processes auto-labeling.
1. The user manually label 100 pieces of data.
2. ClickAI's auto-labeling automatically labels 900 pieces of data.
3. The user inspects the auto-labeled data.
4. Click AI's auto-labeling automatically labels 9000 pieces of data based on the previously labeled 1000 pieces.
5. Repeat step 2-4 as needed.
6. Use labeled data to train model.