Bayesian mode

[nathanin]

to make bayesian inference work

• d-stack the output probabilities with the variances

• add a switch on top of process_tf.process_svs() whether or not to perform this concatenation

• Check for bugs:

  histo_seg/core/process_tf.py

  Lines 104 to 107 in ed21188

  	# if bayesian:
  	# print 'Setting up for bayesian inference mode with {} samples'.format(samples)
  	# process_fn = lambda x: net.bayesian_inference(x, samples=samples)
  	# else:

• Add a flag in the output settings to selectively save variance images - sum variance / pixel, and per-class variance / pixel