How to Deploy Florence-2 with Intel Emerald Rapids
Blog post from Roboflow
Florence-2 is a multimodal vision model developed by Microsoft Research, designed for a variety of foundational vision tasks such as zero-shot object detection, optical character recognition (OCR), and image captioning. This model is compatible with Intel's latest CPU, Emerald Rapids, which offers cost improvements over GPU deployment. Despite being optimized for GPU, Florence-2 can be run on CPUs using Roboflow Inference. The blog post provides a guide on deploying Florence-2 on an Intel Emerald Rapids CPU system via Google Cloud Platform, detailing the provisioning of the CPU instance, installation of necessary software, and execution of inference tasks such as OCR and object detection. Florence-2's performance was benchmarked on several CPU architectures, with Emerald Rapids demonstrating the lowest inference time per image. This makes Florence-2 an attractive option for various computer vision applications, including reading text in images and identifying objects, while also highlighting its versatility and efficiency on different hardware setups.