Microsoft releases Florence 2 Vision models that can outperform larger specialist models

Microsoft has released a set of vision models called Florence 2. Florence 2 is a prompt-based vision model designed for computer vision and image processing tasks such as image description, object recognition, localization, and segmentation. According to Microsoft, Florence 2 can outperform other specialized and larger vision models in some tasks. To train Florence, Microsoft created the FLD-5B dataset, which contains 5.4 billion annotations for 126 million images. The models come in two sizes, with 0.23B and 0.77B parameters, and are available on Hugging Face for commercial use under the MIT license.