aiDoseBoosterTraining

aiDoseBoosterTraining trains a neural network to be used later on in aiDoseBooster. Based on the Noise2Noise idea where noisy movie stack images are used for training.

aiDoseBoosterTraining can be used without a configuration file and all parameters can be provided by command line arguments. For convenience, parameters can also be given using a configuration file with the -u command line argument. Note: If a parameter is passed by file and by command line, the command line overrides the file.

Options are:

CudaDeviceID

The deviceID of the GPU to use.
Argument for command line: -d or –CudaDeviceID
Type: int
Option is mandatory: false
Default value if not set: 0

Input

The file to process.
Argument for command line: -i or –Input
Type: string
Option is mandatory: true

DeadPixelRemoval

Enables removal of dead pixels based on thresholding. If set to ‘relative’, the threshold is X times the STD + MEAN of the image.
Argument for command line: -deadPixel or –DeadPixelRemoval
Type: one of [ABSOLUTE, NONE, RELATIVE]
Option is mandatory: false
Default value if not set: NONE
Possible notations:

  • ABSOLUTE: ABSOLUTE, Absolute, absolute
  • NONE: NONE, None, none
  • RELATIVE: RELATIVE, Relative, relative

DeadPixelThreshold

Threshold above which a pixel is considered invalid.
Argument for command line: -threshold or –DeadPixelThreshold
Type: float
Option is mandatory: true
Only applicable if

  • DeadPixelRemoval = ABSOLUTE or
  • DeadPixelRemoval = RELATIVE

Network

The file to store the trained network configuration.
Argument for command line: -n or –Network
Type: string
Option is mandatory: true

NetworkIn

If given, the training refines the provided network configuration.
Argument for command line: -ni or –NetworkIn
Type: string
Option is mandatory: false
Default value if not set: ’’

Iterations

The number of iterations to train the network.
Argument for command line: -iter or –Iterations
Type: int
Option is mandatory: true

ResetIterations

When refining an existing network, this parameter allows to reset the iteration counter in order to put more weight on the new dataset.
Argument for command line: -resetIter or –ResetIterations
Type: bool
Option is mandatory: false
Default value if not set: false

DeadPixelMap

An image indicating defective pixels (same size as movie frames).
Argument for command line: -deadPixelMap or –DeadPixelMap
Type: string
Option is mandatory: false
Default value if not set: ’’

GainReference

An image for gain reference (same size as movie frames).
Argument for command line: -gainRef or –GainReference
Type: string
Option is mandatory: false
Default value if not set: ’’

Skip

Skip the X first frames of a stack.
Argument for command line: -skip or –Skip
Type: int
Option is mandatory: false
Default value if not set: 0

FlipYAxis

Flip all images on Y axis before any processing.
Argument for command line: -flip or –FlipYAxis
Type: bool
Option is mandatory: false
Default value if not set: false