fbpx

What is Image Annotation?

Share This Post

Share on facebook
Share on linkedin
Share on twitter
Share on email

As mentioned in the previous blog post, Dataset preparation is the first step to create a powerful AI.

Basically, image annotation is adding some information on the image so the AI can analyze them and learn how to reproduce a task.

But for each task that you want your AI to perform, you will need a different annotation type.

Depending on your project you’ll also need a different number of annotated images to achieve great results.

To summarize all that we’ll present to you the different annotation techniques, the number of annotated images needed and the average time spent by images to give you a great overview.

Image segmentation

In order to teach an AI to segment images efficiently, you will need to manually draw a contour around every object on the picture.

Great segmentation AI has been trained on 100 k+ annotated images, in average, people spend 10 minutes by images.

Building an Image segmentation Dataset will often result in a 16 000 h of work!

Object Detection

To teach an AI how to detect objects you need to locate then manually first by creating a bounding box. It’s a rectangle that matches exactly the object.

A great object detector can be created with 20k+ images, on average it takes a minute by images.

Building an Image segmentation Dataset will often result in a 330 h of work!

Face analysis

 

To teach an AI how to recognize a face you need to manually mark some key points on the face. These points will teach the AI how to look up a face to identify a person.

Great face recognition system can be created with 200k+ images, on average it takes 3 minutes by images.

Building an Image segmentation Dataset will often result in a 100 000 h of work!

Image Classification

To teach an AI how to classify images you need to manually them first.

Great images classification system can be created with 200k+ images, on average it takes 5 seconds by images.

Building an Image segmentation Dataset will often result in a 280 h of work!

But fortunately all these annotation task can be optimized by smart annotation tools

You might also like

Get product updates straight in your inbox!

ultricies vulputate, elementum Aenean tristique commodo et,