search

lucasb-eyer / pydensecrf

Python wrapper to Philipp Krähenbühl's dense (fully connected) CRFs with gaussian edge potentials.
MIT License
1.94k stars 414 forks source link

confidence score #23

Closed zhengshou closed 7 years ago

zhengshou commented 7 years ago

Hi,

I am not familiar with denseCRF but I am wondering would it be possible (or make sense) to obtain confidence score for prediction at each pixel from Q ? Thanks.

Best, Zheng

lucasb-eyer commented 7 years ago

So why don't you read the original paper that I'm also referencing in the README? The paper explains exactly what Q is and should answer your question. Q is an approximative distribution, so you could use it as-is, but it's not clear how well-behaved it will be. I recommend running some tests and seeing how Q behaves in your case.

Closing as it's not an actual issue.

zhengshou commented 7 years ago

Thank you and sorry for confusion. I guess my problem actually is not sure how to read values out of Q which is pydensecrf.eigen.MatrixXf object. But now I find can simply convert it as numpy array.

lucasb-eyer commented 7 years ago

Ah, I didn't read it like that, sorry. Yes, it implements the buffer interface so np.array(Q) should just work.