facebookresearch / segment-anything

The repository provides code for running inference with the SegmentAnything Model (SAM), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.
Apache License 2.0
47.38k stars 5.61k forks source link

does sam supports multiple foreground and background points as prompt? #432

Open Vineet-the-git opened 1 year ago

Vineet-the-git commented 1 year ago

I have been trying to notice the change in performance as I change the number of foreground and background points. I am passing a 384 x 384 image to the model. I am passing the coordinates of the model as [[x1,y1],[x2,y2],[x3,y3].....] and corresponding labels [ 0, 1, 1, 1, 0, ...]. But the performance is decreasing as I am increasing the number of points, which is counter-intuitive. Can anyone help me think of a reason why it is happening?

lavishchauhan321 commented 1 year ago

what exactly do you mean by "performance is decreasing", the foreground and background separation mask is getting more far from true mask ?