Multimodal AI
Vision transformer
A type of AI that examines relationships between pieces of a picture to interpret it.
Example
An AI system represents patches of a photograph and learns relationships between them.
Why people use it
It applies a language-inspired network design to learning from pictures.
What you'll hear
“It treats the image as pieces that relate to one another.”
What this means for you
Compare AI systems on the relevant visual task and operating limits.
Can you control it?
Developer-only
The people building or running the AI choose this setup. An everyday user generally needs their help to change how this part works.
Common questions
- Is every vision AI system a vision transformer?
- No. Other designs, including convolutional networks, also process images.
- Do smaller image pieces always improve results?
- No. More pieces add work, and the best choice depends on the task.
- Can it learn from the arrangement of objects?
- Yes. Relationships between parts of an image can help it interpret the scene.