Building segmentation

[lcoandrade]

I'm testing segment-geospatial as an alternative to RasterVision as a semantic segmentation tool. I already have good results with RV and now I would like to know how to determine parameters to make good segmentations of buildings that are comparable to the results I have with RV.

My goal is to use text prompts text_prompt = 'building'.

This is a sample of the results I obtained with:

text_prompt = 'building'
sam.predict(image, text_prompt, box_threshold=0.2, text_threshold=0.2)

[Mo4py]

Test promts are probably not the best promt solution for this task.
Try BBoxes or Points instead.
E.g.: