Maskrcnn-benchmark: How to get data output from inference?

[innerop]

Hey there,

Instead of getting an image that is overlyed with predictions, can we get a JSON with coordinates of the enclosing box and the type of object detected?

That seems like a very basic expectation for automated monitoring of premises and other applications.

Any hints and tips you may have will be appreciated.

[LeviViana]

The model output is actually an object of the type BoxList. You can get the result information by doing something like :

output = model(image)
# Bounding boxes
boxes = output.bbox 
# Labels
labels = output.get_field("labels")
# Scores
scores = output.get_field("scores") 

You can then customize your JSON as you wish.

[fmassa]

Thanks for the explanation @LeviViana !