Compute metrics for segmentations.

stolen and adapted from here: https://sgugger.github.io/how-do-you-find-a-good-learning-rate.html :param num_iters: :param in

:param output_file: This needs to be the full path to the dataset.json you intend to write, so output_file='DATASET_PATH/dataset.json' where

create self.output_folder modify self.output_folder if you are doing cross-validation (one folder per fold) set self.tr_gen

used for if you need to ensemble the five models of a cross-validation. This will restore the model from the checkpoint in fold 0, load all p

THIS DOES NOT TRANSFER SEGMENTATION HEADS!

if epoch is not None we overwrite epoch. Else we use epoch = self.epoch + 1 (maybe_update_lr is called in on_epoch_end which is call

pads shape so that it is divisibly by must_be_divisible_by :param shape: :param must_be_divisible_by: :return:

here we use the standard naming scheme to generate list_of_lists and output_files needed by predict_cases :param model: :param input

data is (metrics x experiments x cases) :param data: :return:

This is a utility function to load any nnUNet trainer from a pkl. It will recursively search nnunet.trainig.network_training for the file tha

if we run with -c then we need to set the correct lr for the first epoch, otherwise it will run the first continued epoch with self.i

We need to wrap this because we need to enforce self.network.do_ds = False for prediction

checking initial vs corrected dataset :param img1: :param img2: :return:

writes a summary.json to folder_with_predictions :param folder_with_gts: folder where the ground truth segmentations are saved. Must be nifti

image must be a color image, so last dimension must be 3. if image is grayscale, tile it first! Segmentation must be label map of same shape

use this instead of get_default_augmentation (drop in replacement) to turn off all data augmentation

Computation of input patch size starts out with the new median shape (in voxels) of a dataset. This is opposed to prior experiments w

per default we use the 50th percentile=median for the target spacing. Higher spacing results in smaller data and thus faster and easi

:param training: :return:

inference_apply_nonlin to sigmoid

TP / (TP + FP)

Use this function to predict a 3D image. It does not matter whether the network is a 2D or 3D U-Net, it will detect that automaticall

:param data: :param do_mirroring: :param mirror_axes: :param use_sliding_window: :param step_size: :p

Used to predict new unseen data. Not used for the preprocessing of the training/test data :param input_files: :return:

data and seg must already have been transposed by transpose_forward. properties are the un-transposed values (spacing etc) :p

TP / (TP + FN)

Set the labels. :param labels= may be a dictionary (int->str), a set (of ints), a tuple (of ints) or a list (of ints). Labels will onl

- we increase roation angle from [-15, 15] to [-30, 30] - scale range is now (0.7, 1.4), was (0.85, 1.25) - we don't do elast

TN / (TN + FP)

used to swap between folds for inference (ensemble of models from cross-validation) DO NOT USE DURING TRAINING AS THIS WILL NOT UPDAT