Plots confusion matrix Args: normalize (bool, optional): whether to normalize column values. Defaults to True. save_d

Loop through the registered actions and fire all callbacks Args: hook The name of the hook to check, defaults to all

CSP Bottleneck with 3 convolutions. CSP Bottleneck https://github.com/WongKinYiu/CrossStagePartialNetworks Args: c1 (int)

This is the dataloader used in the training process The same as that of 2D YOLO, just built around a different Dataset definition Args:

Loads an ensemble of models weights=[a,b,c] or a single model weights=[a] or weights=a Called during validation. Args: weights (str o

Return intersection-over-union (Jaccard index) of boxes. Both sets of boxes are expected to be in (z1, x1, y1, z2, x2, y2) format. https://git

Rescale bounding box coordinates (zxyzxy) from img1_shape to img0_shape. Args: img1_shape (Tuple[int]): size of the resized image that pr

Convert nx6 boxes from [z1, x1, y1, z2, x2, y2] to [z, x, y, d, w, h] where zxy1=top-left, zxy2=bottom-right Args: labels (torch.tensor o

Resizes a 3D tensor to a cube with edge length new_size. Also adds the channel dimension. Args: nifti_tensor (torch.Tensor): The tens

In-place clip zxyzxy bounding boxes to image shape (depth, height, width) Args: boxes (torch.tensor or np.ndarray): bounding boxes to cli

Defines label paths as a function of the image paths. Filters for .nii and .nii.gz files. Args: img_paths (List[str]): list of image fil

Runs Non-Maximum Suppression (NMS) on inference results for 3D bounding box predictions. Args: prediction (torch.tensor): tensor of predi

Reads a nifti file and converts it to a torch tensor Args: filepath (str): Path to the nifti file Returns: nifti (torch.tens

Decorator to make all processes in distributed training wait for each local_master to do something.

Reshapes the tensor from height, width, depth order to depth, height, width to make it compatible with torch convolutions. Args: nift

Convert nx6 boxes from [z, x, y, d, w, h] to [z1, x1, y1, z2, x2, y2] where zxy1=top-left, zxy2=bottom-right Args: labels (torch.tensor o

Performs non-maximum suppression (NMS) on bounding boxes according to their intersection-over-union (IoU). NMS iteratively removes lowe

Compute the average precision, given the recall and precision curves. Source: https://github.com/rafaelpadilla/Object-Detection-Metrics. # Ar

Measures the distance between the centers of two bounding boxes Args: box1 (torch.Tensor): First bounding box to calculate distance for.

Returns the IoU of box1 to box2. box1 is 6, box2 is nx6. Args: box1 (torch.Tensor): First bounding box to calculate IoU for.

Returns the intersection over box2 volume given box1, box2. Boxes are z1x1y1z2x2y2 box1: np.array of shape(6) box2: np.array of s

Regression formula to convert model output into bounding box predictions.

Build targets for compute_loss() input targets with format (image,class,z,x,y,d,w,h) Args: pred (torch.Tensor): Examp

Caches dataset labels, verifies images and reads their shapes. See: verify_image_label function Args: path (pathlib.Path,

Compute the average precision, given the recall and precision curves # Arguments recall: The recall curve (list) precision: Th

Takes a parent directory and looks for files within those directories. Output organized to fit YOLO inference requirements. Args: par

Takes a parent directory or list of parent directories and looks for files within those directories. Output organized to fit YOLO training re

Fuse convolution and batchnorm layers https://tehnokv.com/posts/fusing-batchnorm-and-conv/ Args: conv (torch.Module): Conv layer to fuse.

Initializes parameters for layers that need extra parameters set. Args: model (torch.Module): YOLO model to initialize weights for

Reads a nifti file, converts it to a torch.tensor, and reshapes and resizes it for use in the YOLO3D model. Args: i (int): Datase

Makes nifti masks out of YOLO label txt files. Only works for one label per mask. Labels should have one prediction without confidence metri

Model information. Args: model (torch.Module): Model to get information from verbose (bool, optional): Whether or not to print na

Makes nifti masks out of a YOLO label txt file. Saves highest confidence mask for each class. Args: bbox_path: path to the YOLO labe

Checks anchor fit to data and recomputes if necessary. Args: dataset (torch.Dataset): Dataset the anchors will be used with. mode

Creates kmeans-evolved anchors from training dataset using niftis Arguments: dataset (str or torch.Dataset): path to dat

Normalizes 3D CTs in Hounsfield Units (+/- 1024) to within 0 and 1. Args: imgs (torch.tensor): unnormalized model input Returns:

Volume normalizes 3D MR images to mean 0 and standard deviation 1. Args: imgs (torch.tensor): unnormalized model input Returns:

Configures the model using hyperparameters from the model yaml file. Note: model yaml file (e.g. yolov3Ds.yaml) is distinct from the training

Return correct predictions matrix. Both sets of boxes are in (z1, x1, y1, z2, x2, y2) format. Arguments: detections (Array[N, 8]), z1

Return intersection-over-union (Jaccard index) of boxes. Both sets of boxes are expected to be in (z1, x1, y1, z2, x2, y2) format.

Randomly zooms in or out of a random part of the input image. Args: im (torch.Tensor): 3-D tensor to be augmented. labels (List[f

Reduce loss as specified. Args: loss (Tensor): Elementwise loss tensor. reduction (str): Options are "none", "mean" and "sum".

Creates nifti masks for MedYOLO bounding boxes found in bbox_dir that have a corresponding nifti image in nifti_dir. See mask maker functions