How SAM generates segmentation annotations from points?

[WangRongsheng]

Hi Viet Anh,

I would like to ask you some questions. When I use SAM to do segmentation annotation, I use the point pattern. How does SAM determine the split area based on the points I enter?

Looking forward to your answer!

[vietanhdev]

The SAM is a deep learning model. The decoder receives 2 inputs:

• The embedding, which is the output of the encoder.
• The prompts (points: coordinates x, y of point, the box (coordinates x, y of box), and a mask). It (the decoder) will run as a deep-learning image segmentation model to produce binary masks. After that, I use contour finding algorithm in OpenCV with the output (the first binary mask) to obtain the object boundaries.

[WangRongsheng]

Can you give some usage demo code? Thank you @vietanhdev