dataset package

dataset.dataset_utils module

dataset.dataset_utils.calculateAngles(nAngles: int) → tensor

Return array of angles in [0, 360) degrees.

Args:

nAngles (int): Number of desired angles.

Returns:

angles (tensor): Uniformly spaced angles, excluding 360 degrees.

dataset.dataset_utils.decimateAngles(nAnglesFull: int, downsample: int) → tensor

Decimate an array of angles.

Args:

• nAnglesFull (int): Number of angles at full resolution.

• downsample (int): Downsampling factor.

Returns:

angles (tensor): Array of downsampled angles.

Example:

As an example, suppose the full set contains 451 angles. Below are various downsampling factors.

• 4x : 113 views

• 8x : 57 views

• 16x : 29 views

• 32x : 15 views

dataset.dataset_utils.decimateSinograms(sinograms: tensor, downsample: int)

Decimate the sinograms by angle.

Args:

• sinograms (tensor): Dimensions are

[batchSize, nChannels, nRows (nAngles), nCols].

• downsample (int): Downsampling factor

Returns:

Decimated sinogram tensor.

dataset.dataset_utils.getMask(imgShape: List[int])

Return a mask of an inscribed circle in the image.

Args:

imgShape (List[int]): [nRows, nCols]

Returns:

Mask of 1 inside circle and 0 outside circle

dataset.dataset_utils.resizeSinograms(sinograms: array, nRows: int = 128)

Reize projection images to (nRows, nRows).

Args:

• sinograms (array): Dimensions are

[nRows, nAngles, nRows (=nColumns)]

• nRows (int): Desired image size

Returns:

Resized sinogram

dataset.dataset_utils.sampleSinograms(sinograms: tensor, rowRange: List[int])

Calculate train or test set from a subset of rows.

Args:

• sinograms (tensor): Sampled row-by-row.

Dimensions are [batchSize, nAngles, nColumns]. * rowRange (List[int]): Row range to sample. Dimensions are [startRow, endRow].

Returns:

Sampled sinogram tensor.

dataset.metrics module

dataset.metrics.calculateMSE(gt: tensor, img: tensor) → float

Calculate mean squared error.

Args:

• gt (tensor): Ground truth tensor.

• img (tensor): Prediction. Same size as gt.

Returns:

Mean squared error (float).

dataset.metrics.calculatePSNR(gt: tensor, img: tensor) → float

Calculate peak signal to noise ratio.

Args:

• gt (tensor): Ground truth tensor.

• img (tensor): Prediction. Same size as gt.

Returns:

Peak signal to noise ratio (float).

dataset.metrics.calculateSSIM(gt: tensor, img: tensor) → float

Calculate structural similarity image metric.

Args:

• gt (tensor): Ground truth tensor.

• img (tensor): Prediction. Same size as gt.

Returns:

Structural similarity image metric (float).

dataset.projections module

dataset.projections.loadSinograms(projDir: str) → tensor

Return array of 2D sinograms from row-by-row projections

Args:

projDir: Directory with .tif files

Returns:

• sinograms (tensor):

(batchSize = nRows) x (nChannels = 1) x nAngles x nColumns

Notes:

For absorption and dark field, take the neg-log of each projection. For differential or integrated phase, negate each projection to make most values positive.

dataset.projections.saveMultipleExperiments(projDirs: List[str], trainSinoFile: str, testSinoFile: str)

Load projections, and save them in train and test sets. Uses the top half of each projection for training and the bottom half for testing.

Args:

• projDirs (List[str]): Directories with tif projections.

• trainSinoFile (str): Filename to save train projections.

• testSinoFile (str): Filename to save test projections.

Returns:

Saves projections in two files for training and testing.