Segment Anything Model (SAM) is an image segmentation technology developed by Meta AI. It stands out for its ability to apply, without additional training, to unknown objects and images. Here's an overview of our experimentation with this promising technology.
Setup and installation
SAM is easy to configure and can run on an M1 Mac thanks to its Python implementation using Facebook Research's GitHub repository. SAM is also an open-source project, making it freely accessible to users.
Experimentation
The main objective of our experimentation was to detect boxes and parts on a mechanical drawing. SAM was used to extract all possible boxes from a plan, being able to generate precise masks for each detected object. Florence-2 was subsequently used to understand the plan's content.
