Image Understanding
Definition
Image understanding reveals content and relationships in image data for a specific task.
Image understanding here refers to task-related evaluation of image content by AI. Visual question answering is a research example: a system processes an image and a related question and generates an answer.
Simply explained
Like a search task in a 'Where's Waldo?' picture. The system looks for computational clues; this does not imply it experiences or understands a scene like a human.
Why it matters in practice
In an image project, the actual questions to be answered should be defined instead of a general understanding.
Example from business
Possible example: A team checks whether a model can answer questions about recognizable objects in their own product photos.
Limits and pitfalls
- A general image description does not cover every specific question. VQA work distinguishes targeted questions about details and context from general image captions.
Origin & history
The cited 2016 VQA version describes open questions and answers about images as a task to investigate visual understanding.
Use in mid-sized companies
- Possible application areas include image archives and internal documentation processes, each with a professionally evaluated catalog of specific image questions.
Frequently asked questions
Is this the same as an image caption?
No, specific questions may require different details and relationships.
Is human consciousness claimed here?
No, it means processing defined image tasks.
Sources
- VQA: Visual Question Answering · Agrawal et al. (2016): VQA: Visual Question Answering
Editorially reviewed on: