Meta AI has released EUPE, a compact “Vision Encoder” model family with fewer than 100M parameters that targets strong performance on multiple vision tasks.
The project claims results that rival specialist models across image understanding, dense prediction, and vision-language-model (VLM) related tasks.
EUPE is published via the official Facebook Research GitHub repository, enabling researchers and developers to access code and model resources.
The release is positioned as a practical alternative to larger or specialized vision encoders, potentially improving efficiency for downstream applications.
A Compact Vision Encoder Family Under 100M Parameters That Rivals Specialist Models Across Image Understanding, Dense Prediction, and VLM Tasks