Decoder. plain-English AI glossary

Image Encoder

● Core

Neural network component that converts images into dense vector representations so images can be compared and retrieved.

Think of it like

Reducing a photo to its essence—color palette, composition, key objects—rather than storing every pixel.

Example

ResNet outputs 2048-d embeddings; two cat photos produce very similar embeddings even if one is closer-up; dissimilar images are far apart in embedding space.

How it actually works

Vision models like ViT or CLIP produce embeddings. Embeddings trained to be similar for semantically related images (contrastive learning with image-image pairs or image-text pairs). Can be used for image search, deduplication, or as feature input to downstream models.

For product teams

Powers visual search and similarity-based recommendations.

For engineers

Forward pass of image through network backbone up to penultimate layer; output is a 512-2048-d vector; stored in vector DB for search.

VLM — Component of.
Powers.

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open Decoder Add to Chrome