The Mobile Segment Anything Model is constructed upon the foundation of the Facebook SAM model. It serves as a lightweight iteration of SAM, designed to empower edge devices to seamlessly execute this model on their CPUs. The model's publication is rooted in the research paper titled Mobile Segment Anything, authored by Chaoning Zhang, Dongshen Han, Yu Qiao, Jung Uk Kim, Sung-Ho Bae, Seungkyu Lee, and Choong Seon Hong.
Tutorial Created by Subin Erattakulangara
The MobileSAM model empowers users to effortlessly segment entities within images through a simple click. This lightweight variant of the SAM, aptly named MobileSAM, boasts a remarkable reduction in size, being more than 60 times smaller while maintaining performance levels equivalent to the original SAM.
In terms of inference speed, MobileSAM demonstrates remarkable efficiency. On a single GPU, it processes images in approximately 10 milliseconds per image, with 8 milliseconds allocated to the image encoder and 4 milliseconds to the mask decoder.
- Navigate to the DSA module
runCustomAIModel.
-
Select "Mobile Segment Anything" from the dropdown menu for AI models.
-
Identify the nuclei you wish to select, and then simply click "Submit." Alternatively, you can continuously select nuclei by utilizing the dropdown menu, as demonstrated in the GIF below.
