ALIGN

What it does

ALIGN is a vision-language model from Google Research, described at ICML 2021, that learns matching image and text embeddings with a dual encoder trained by contrastive loss. It was trained on over a billion noisy image alt-text pairs with little filtering, showing that data scale can make up for noise. Its embeddings support image-text retrieval and zero-shot image classification.

Compare with

• vsCohere Embed
• vsOpenAI Embeddings
• vsNV-Embed

People who use it also use

Not enough marks yet to tell. When people who use ALIGN mark the rest of their stack, it shows up here.