A vision foundation model is a broadly trained model that learns reusable visual representations from large amounts of image or visual data.
A vision foundation model learns broad visual patterns that can be reused across many visual tasks.
It combines broad pre-training, reusable visual capability, and adaptation to tasks such as classification, detection, or image understanding. It is not simply a model that generates images.
Someone who first studies a huge visual library before specializing in tasks such as classification, detection, or image understanding.
Assign a label to an image, such as identifying a product category.
Connect visual content with language, such as answering a question about an image.
Find visually similar products, documents, or scenes.
Locate objects or separate regions when the model supports those tasks.
Vision foundation model: Learns reusable visual representations for many tasks.
Image generator: Creates new images from an input or prompt.
Vision foundation model: A broadly trained model.
Computer vision: The wider field of methods and systems for visual data.
Vision foundation model: Primarily learns visual representations.
Multimodal model: Connects or reasons across multiple modalities.
Explore next
A vision foundation model learns broad visual representations that can support many downstream visual tasks.
A video will be added when a suitable visual explainer is available.