Module qute.transforms.debug
Debugging transforms.
Classes
class DebugCheckAndFixAffineDimensions (expected_voxel_size, *args, **kwargs)-
Check that the affine transform is (4 x 4).
Constructor.
Parameters
name:str- Name of the Informer. It is printed as a prefix to the output.
Expand source code
class DebugCheckAndFixAffineDimensions(Transform): """Check that the affine transform is (4 x 4).""" def __init__(self, expected_voxel_size, *args, **kwargs): """Constructor. Parameters ---------- name: str Name of the Informer. It is printed as a prefix to the output. """ super().__init__() self.name = "" if "name" in kwargs: self.name = kwargs["name"] self.expected_voxel_size = expected_voxel_size def __call__(self, data): """Call the Transform.""" prefix = f"{self.name} :: " if self.name != "" else "" if type(data) is list: for item in data: if type(item) is dict: for key in item.keys(): sub_item = item[key] if type(sub_item) is MetaTensor and hasattr(sub_item, "affine"): if sub_item.affine.shape != (4, 4): print( f"{prefix}Affine matrix needs correcting! Current shape is {sub_item.affine.shape}" ) sub_item.affine = torch.tensor( [ [self.expected_voxel_size[0], 0.0, 0.0, 0.0], [0.0, self.expected_voxel_size[1], 0.0, 0.0], [0.0, 0.0, self.expected_voxel_size[2], 0.0], [0.0, 0.0, 0.0, 1.0], ] ) elif ( sub_item.affine[0, 0] != self.expected_voxel_size[0] or sub_item.affine[1, 1] != self.expected_voxel_size[1] or sub_item.affine[2, 2] != self.expected_voxel_size[2] or sub_item.affine[0, 3] != 0.0 ): print( f"{prefix}Affine matrix needs correcting! Current matrix is {item.affine}" ) sub_item.affine = torch.tensor( [ [self.expected_voxel_size[0], 0.0, 0.0, 0.0], [0.0, self.expected_voxel_size[1], 0.0, 0.0], [0.0, 0.0, self.expected_voxel_size[2], 0.0], [0.0, 0.0, 0.0, 1.0], ] ) else: # The affine matrix is fine, nothing to correct return data if type(data) is dict: for key in data.keys(): item = data[key] if type(item) is MetaTensor and hasattr(item, "affine"): if item.affine.shape != (4, 4): print( f"{prefix}Affine matrix needs correcting! Current shape is {item.affine.shape}" ) item.affine = torch.tensor( [ [self.expected_voxel_size[0], 0.0, 0.0, 0.0], [0.0, self.expected_voxel_size[1], 0.0, 0.0], [0.0, 0.0, self.expected_voxel_size[2], 0.0], [0.0, 0.0, 0.0, 1.0], ] ) elif ( item.affine[0, 0] != self.expected_voxel_size[0] or item.affine[1, 1] != self.expected_voxel_size[1] or item.affine[2, 2] != self.expected_voxel_size[2] or item.affine[0, 3] != 0.0 ): print( f"{prefix}Affine matrix needs correcting! Current matrix is {item.affine}" ) item.affine = torch.tensor( [ [self.expected_voxel_size[0], 0.0, 0.0, 0.0], [0.0, self.expected_voxel_size[1], 0.0, 0.0], [0.0, 0.0, self.expected_voxel_size[2], 0.0], [0.0, 0.0, 0.0, 1.0], ] ) else: # The affine matrix is fine, nothing to correct pass elif type(data) is MetaTensor and hasattr(data, "affine"): if data.affine.shape != (4, 4): print( f"{prefix}Affine matrix needs correcting! Current shape is {data.affine.shape}" ) data.affine = torch.tensor( [ [self.expected_voxel_size[0], 0.0, 0.0, 0.0], [0.0, self.expected_voxel_size[1], 0.0, 0.0], [0.0, 0.0, self.expected_voxel_size[2], 0.0], [0.0, 0.0, 0.0, 1.0], ] ) elif ( data.affine[0, 0] != self.expected_voxel_size[0] or data.affine[1, 1] != self.expected_voxel_size[1] or data.affine[2, 2] != self.expected_voxel_size[2] or data.affine[0, 3] != 0.0 ): print( f"{prefix}Affine matrix needs correcting! Current matrix is {data.affine}" ) data.affine = torch.tensor( [ [self.expected_voxel_size[0], 0.0, 0.0, 0.0], [0.0, self.expected_voxel_size[1], 0.0, 0.0], [0.0, 0.0, self.expected_voxel_size[2], 0.0], [0.0, 0.0, 0.0, 1.0], ] ) else: # The affine matrix is fine, nothing to correct pass return dataAncestors
- monai.transforms.transform.Transform
- abc.ABC
class DebugExtractChannel (channel_num: int, mask: bool = False, mask_threshold: float = 0.01, to_binary: bool = False)-
Extract just one channel from a multi-channel tensor.
Constructor.
Parameters
channel_num:int- Index of channel to be extracted.
mask:bool = False- Set to Ture to mask the channel by its binarized version. It
mask_threshold:float = 1e-3- Threshold for the masked image.
to_binary:bool = False- Set to True to convert to binary via sigmoid transform and thresholding. It implies to_binary = True and will ignore the passed argument for it..
Expand source code
class DebugExtractChannel(Transform): """ Extract just one channel from a multi-channel tensor. """ def __init__( self, channel_num: int, mask: bool = False, mask_threshold: float = 1e-2, to_binary: bool = False, ): """Constructor. Parameters ---------- channel_num: int Index of channel to be extracted. mask: bool = False Set to Ture to mask the channel by its binarized version. It mask_threshold: float = 1e-3 Threshold for the masked image. to_binary: bool = False Set to True to convert to binary via sigmoid transform and thresholding. It implies to_binary = True and will ignore the passed argument for it.. """ super().__init__() self.channel_num = channel_num self.mask = mask self.mask_threshold = mask_threshold self.to_binary = to_binary def __call__(self, data): """Call the Transform.""" # Check the number of dimensions in output tensor to determine if it's 2D or 3D if len(data.shape) not in [4, 5]: raise ValueError( "Unexpected number of dimensions, expected 4 (2D) or 5 (3D)." ) # Determine if the data is 2D or 3D and adjust indexing accordingly dim_idx = (slice(None),) * (len(data.shape) - 2) # Extract the selected channel out = data[:, self.channel_num, None, Unpack[dim_idx]] # Binarize if requested bw = 1.0 if self.to_binary or self.mask: bw = torch.sigmoid(out) > 0.5 # Mask if requested if self.mask: out = bw * out out[out <= self.mask_threshold] = 0 return outAncestors
- monai.transforms.transform.Transform
- abc.ABC
class DebugInformer (*args, **kwargs)-
Simple reporter to be added to a Composed list of Transforms to return some information. The data is returned untouched.
Constructor.
Parameters
name:str- Name of the Informer. It is printed as a prefix to the output.
Expand source code
class DebugInformer(Transform): """ Simple reporter to be added to a Composed list of Transforms to return some information. The data is returned untouched. """ def __init__(self, *args, **kwargs): """Constructor. Parameters ---------- name: str Name of the Informer. It is printed as a prefix to the output. """ super().__init__() self.name = "" if "name" in kwargs: self.name = kwargs["name"] def __call__(self, data): """Call the Transform.""" prefix = f"{self.name} :: " if self.name != "" else "" if type(data) is tuple and len(data) > 1: data = data[0] if type(data) is torch.Tensor: print( f"{prefix}" f"Type = Torch Tensor: " f"size = {data.size()}, " f"type = {data.dtype}, " f"min = {data.min()}, " f"mean = {data.double().mean()}, " f"median = {torch.median(data).item()}, " f"max = {data.max()}" ) elif type(data) is np.ndarray: print( f"{prefix}" f"Type = Numpy Array: " f"size = {data.shape}, " f"type = {data.dtype}, " f"min = {data.min()}, " f"mean = {data.mean()}, " f"median = {np.median(data)}, " f"max = {data.max()}" ) elif type(data) is str: print(f"{prefix}String: value = '{str}'") elif str(type(data)).startswith("<class 'itk."): # This is a bit of a hack... data_numpy = np.array(data) print( f"{prefix}" f"Type = ITK Image: " f"size = {data_numpy.shape}, " f"type = {data_numpy.dtype}, " f"min = {data_numpy.min()}, " f"mean = {data_numpy.mean()}, " f"median = {np.median(data_numpy)}, " f"max = {data_numpy.max()}" ) elif type(data) is dict: # A dictionary, most likely of "image" and "label" print(f"{prefix}Dictionary with keys: ", end=" ") for key in data.keys(): value = data[key] t = type(value) if t is MetaTensor: print( f"'{key}': shape={data[key].shape}, dtype={data[key].dtype};", end=" ", ) if hasattr(data[key], "meta") and "affine" in data[key].meta: a = data[key].meta["affine"] print( f"voxel size=({a[0, 0]}, {a[1, 1]}, {a[2, 2]});", end=" ", ) elif t is dict: print(f"'{key}': dict;", end=" ") elif t is pathlib.PosixPath: print(f"'{key}': path={str(data[key])};", end=" ") else: print(f"'{key}': {t};", end=" ") print() elif type(data) is MetaTensor: print(f"{prefix}MONAI MetaTensor: shape={data.shape}, dtype={data.dtype}") else: try: print(f"{prefix}{type(data)}: {str(data)}") except: print(f"{prefix}Unknown type!") return dataAncestors
- monai.transforms.transform.Transform
- abc.ABC
class DebugMinNumVoxelCheckerd (keys: tuple, class_num: int, min_fraction: float = 0.0)-
Simple reporter to be added to a Composed list of Transforms to return some information. The data is returned untouched.
Constructor.
Parameters
class_num:int- Number of the class to check for presence.
Expand source code
class DebugMinNumVoxelCheckerd(MapTransform): """ Simple reporter to be added to a Composed list of Transforms to return some information. The data is returned untouched. """ def __init__(self, keys: tuple, class_num: int, min_fraction: float = 0.0): """Constructor. Parameters ---------- class_num: int Number of the class to check for presence. """ super().__init__(keys=keys) self.keys = keys self.class_num = class_num self.min_fraction = min_fraction def __call__(self, data): """Call the Transform.""" # Make a copy of the input dictionary d = dict(data) # Get the keys keys = d.keys() # Only work on the expected keys for key in keys: if key not in self.keys: continue # The Tensor should be in OneHot format, and the class number should # map to the index of the first dimension. if self.class_num > (d[key].shape[0] - 1): raise Exception("`num_class` is out of boundaries.") num_voxels = d[key][self.class_num].sum() if self.min_fraction == 0.0: if num_voxels > 0: return d if self.min_fraction > 0.0: area = d[key].shape[1] * d[key].shape[2] if num_voxels / area >= self.min_fraction: return d raise Exception( f"The number of voxels for {self.class_num} is lower than expected ({num_voxels})." ) # Return data return dAncestors
- monai.transforms.transform.MapTransform
- monai.transforms.transform.Transform
- abc.ABC