From 4eb1f5d3181a5d95d39bd207cd932a3613b7dffe Mon Sep 17 00:00:00 2001 From: Fabian Isensee Date: Wed, 5 Jun 2024 15:17:23 +0200 Subject: [PATCH] fix a bug where the spacing would not be transposed in export --- nnunetv2/evaluation/evaluate_predictions.py | 1 - nnunetv2/inference/export_prediction.py | 10 ++++++---- 2 files changed, 6 insertions(+), 5 deletions(-) diff --git a/nnunetv2/evaluation/evaluate_predictions.py b/nnunetv2/evaluation/evaluate_predictions.py index a7af531e4..1ecbf3255 100644 --- a/nnunetv2/evaluation/evaluate_predictions.py +++ b/nnunetv2/evaluation/evaluate_predictions.py @@ -92,7 +92,6 @@ def compute_metrics(reference_file: str, prediction_file: str, image_reader_writ # load images seg_ref, seg_ref_dict = image_reader_writer.read_seg(reference_file) seg_pred, seg_pred_dict = image_reader_writer.read_seg(prediction_file) - # spacing = seg_ref_dict['spacing'] ignore_mask = seg_ref == ignore_label if ignore_label is not None else None diff --git a/nnunetv2/inference/export_prediction.py b/nnunetv2/inference/export_prediction.py index 33035676b..f5cdb958d 100644 --- a/nnunetv2/inference/export_prediction.py +++ b/nnunetv2/inference/export_prediction.py @@ -23,14 +23,15 @@ def convert_predicted_logits_to_segmentation_with_correct_shape(predicted_logits torch.set_num_threads(num_threads_torch) # resample to original shape + spacing_transposed = [properties_dict['spacing'][i] for i in plans_manager.transpose_forward] current_spacing = configuration_manager.spacing if \ len(configuration_manager.spacing) == \ len(properties_dict['shape_after_cropping_and_before_resampling']) else \ - [properties_dict['spacing'][0], *configuration_manager.spacing] + [spacing_transposed[0], *configuration_manager.spacing] predicted_logits = configuration_manager.resampling_fn_probabilities(predicted_logits, properties_dict['shape_after_cropping_and_before_resampling'], current_spacing, - properties_dict['spacing']) + [properties_dict['spacing'][i] for i in plans_manager.transpose_forward]) # return value of resampling_fn_probabilities can be ndarray or Tensor but that does not matter because # apply_inference_nonlin will convert to torch predicted_probabilities = label_manager.apply_inference_nonlin(predicted_logits) @@ -123,13 +124,14 @@ def resample_and_save(predicted: Union[torch.Tensor, np.ndarray], target_shape: if isinstance(dataset_json_dict_or_file, str): dataset_json_dict_or_file = load_json(dataset_json_dict_or_file) + spacing_transposed = [properties_dict['spacing'][i] for i in plans_manager.transpose_forward] # resample to original shape current_spacing = configuration_manager.spacing if \ len(configuration_manager.spacing) == len(properties_dict['shape_after_cropping_and_before_resampling']) else \ - [properties_dict['spacing'][0], *configuration_manager.spacing] + [spacing_transposed[0], *configuration_manager.spacing] target_spacing = configuration_manager.spacing if len(configuration_manager.spacing) == \ len(properties_dict['shape_after_cropping_and_before_resampling']) else \ - [properties_dict['spacing'][0], *configuration_manager.spacing] + [spacing_transposed[0], *configuration_manager.spacing] predicted_array_or_file = configuration_manager.resampling_fn_probabilities(predicted, target_shape, current_spacing,