Imagen

tags

Transformers, Diffusion models, Computer vision, NLP, T5, CLIP

paper

(Saharia et al. 2022)

Architecture

This is based on the U-net diffusion architecture with a few extensions. T5 or CLIP or BERT is used as a frozen text encoder.

Parameter count

2B

Bibliography

1. Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily Denton, Seyed Kamyar Seyed Ghasemipour, et al.. May 23, 2022. "Photorealistic Text-to-image Diffusion Models with Deep Language Understanding". arXiv. DOI.