computer vision - Image labeling performance using CRF -


i need develop image labeling application, task i'm considering using conditional random fields (crf) on set of superpixels, there exists quite few papers point out technology state of art task. usual task devided 2 tasks:

  • training model: problem obtaining parameter vector 'w', using example
  • testing: obtaining feasible label assignment of given set of superpixels, i.e argmax(p(y|x))

i'm aware of training-time quite high, have not found testing-time nor performance, have , idea of how time take testing problem? suppose depend on number of labels, image size, implementation, hardware, etc

testing slowish because still have solve graph cuts problem (but nothing training). there implementation can try out @ http://drwn.anu.edu.au/drwnprojmultiseg.html (you have seen stephen gould's papers).

i still have log file. bit hard interpret following may not totally accurate. on super fast machine, think took about:

  • 4.5 hours cpu time train 20 classes on 276 images msrc dataset
  • 50 mins cpu time classify 256 images, of spent doing alpha expansion

Comments

Popular posts from this blog

linux - xterm copying to CLIPBOARD using copy-selection causes automatic updating of CLIPBOARD upon mouse selection -

qt - Errors in generated MOC files for QT5 from cmake -