Segmenting multiple object instance without prompt

[gmnarciso]

Hi! Thank you very much for sharing your code! I was able to run it on my local PC. I have also tried it on my own dataset, particularly for tree crown delineation on UAV images. While it worked, it however fails to segment other trees within the image. It only segments the object within the input prompt or bounding box. In this regard, how do I customize this in a way that it is able to segment all instances of the object in the image even only using a single prompt? Moreover, is also impossible to make this code to segment the objects within the image without a prompt same way how other CNN works

[Ultraman6]

you may see SamAutomaticMaskGenerator inference api or the work of other called SAM-SP, the former construct prompt all around layers with boxes,the latter explore a training mode with end-to-end inference with SAM