What is ImageIdentify?
ImageIdentify is a function in Wolfram Language that identifies the content of an image. This site showcases how to use it.
General
What is Wolfram Language?
Wolfram Language is a powerful computer language with extensive knowledge of computation and real-world data built in. It's the core technology developed by Wolfram Research over nearly 30 years and is the foundation of well-known products such as Mathematica and Wolfram|Alpha.
Is this site an example of artificial intelligence?
Yes. Image identification has long been a prominent problem in the field of artificial intelligence. And it's something computers have traditionally not been able to do at all well.
Is Wolfram Research an artificial intelligence company?
Wolfram Research has been a leader in developing and deploying computational intelligence for nearly 30 years. The company has built into its products solutions to many problems that were traditionally considered to be in the domain of artificial intelligence.
How can I find out more about the background to the site?
Check out Stephen Wolfram's announcement blog post.
Identifying Images
What can ImageIdentify recognize?
It's trained to recognize about 10,000 common kinds of objects.
What can't it recognize?
UFOs. Abstract art. Things that aren't real everyday objects (except for a few notable fictional entities). Specific people (except for a few notable ones).
What if there is more than one object in the picture?
ImageIdentify tries to identify the "main object" in a picture. (There will soon be other powerful Wolfram Language functions for dealing with multiple objects.)
How can I help it recognize something?
Make sure the object is in its "natural habitat." Like humans, ImageIdentify needs to use contextual clues. It'll do much better recognizing a boat if it can see that it's on water—rather than, say, floating in space.
What is going on when it gets the completely wrong result?
Most often it's because it learned the wrong lesson from training images. Say your picture is of a tiny party hat. It's seen such a hat before in its training, but only in pictures that are identified as being of chihuahuas. So it'll say the hat "is" a chihuahua.
Does the site learn from its users?
Yes. Feedback from users is crucial in helping ImageIdentify improve and in filling in gaps in its knowledge. The statistical distribution of types of images that people provide also gives us important data.
Using the Site
Does the site work on phones?
Yes. You can give it pictures from your photo gallery or directly from your phone camera.
On the desktop, can I repeatedly drag and drop pictures?
Yes. After you get a result, you can drag another picture on top of where the previous one appears.
Can I drag pictures from any webpage?
Mostly yes, but there are exceptions. For example, you may not be able to drag pictures between different web browsers. And some sites may display their images in ways that do not allow drag and drop.
What image formats can I use?
You can use pretty much any standard format, including JPEG, GIF, PNG or any of the many formats supported by Wolfram Language.
Does the site work faster on some images than others?
Images that have fewer pixels will be faster to upload and process. What's in the image won't affect the speed of the recognition process.
Are the pictures I upload stored?
A thumbnail is stored so that we can give you a page you can share. In addition, we are collecting anonymous images to help train future versions of ImageIdentify. (See also our general Privacy Policy.)
Where to Go from Here
Is there an API?
Yes! The Wolfram Cloud lets you create an instant API that uses ImageIdentify however you want.
Can I use ImageIdentify locally on my computer?
Yes. It's part of Wolfram Language, which runs on standard desktop, server and embedded computer systems. ImageIdentify does not require internet access.
Can I set up image identification for specialized types of images?
Yes. Wolfram Language has powerful built-in machine learning functions for doing exactly this.
What else does Wolfram Language do?
Lots! ImageIdentify is just one of about 5,000 built-in functions in Wolfram Language.