Image Captioning
Generating a natural language description of an image—what objects, actions, and relationships appear in the photo.
Think of it like
Writing an alt-text for an image that sighted users cannot see.
Example
"A person hiking in the mountains with a sunset in the background" from a photo of a hiker silhouetted against dusk.
How it actually works
Combines vision encoder (CNN or ViT) with sequence generation (LSTM or transformer decoder). Training pairs images with human-written captions. Generated captions are often generic or miss fine details. Beam search during inference trades speed for quality.
For product teams
Accessibility feature and enables search by description.
For engineers
Image encoder outputs embeddings; decoder attends to embeddings and generates words token-by-token; beam search finds top-k captions.
Related
- VLM — Variant of.
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 DecoderAdd to Chrome