PR: Enhanced Inference and Evaluation Capabilities for nnU-Net Models
This PR introduces enhancements to the inference and evaluation processes of nnU-Net models, specifically targeting the functionality to execute these processes on individual folds in addition to the existing ensembling approach. Below is a summary of the key changes incorporated into this update:
Changes
Folds Selection for Inference: A new --folds argument has been added to the run_inference.py script, enabling the specification of which folds to use during inference. This feature allows for greater flexibility, as users can now choose to run inference on individual folds or continue using all folds by default, as was previously the case.
Commit: 942babf82628c880b9f126234872008e28fa0f6d
Automatic Directory Creation for Evaluation Output: The run_evaluation.py script has been updated to automatically create the directory for storing the output .csv file if it does not already exist. This enhancement simplifies the evaluation process by reducing the need for manual directory management.
Commit: ac8030891948a2a20c5bcde1b8baeae945eefdc7
Enhanced Evaluation Script: Improvements have been made to the inference_and_evaluation.sh script to support evaluation not only on the ensemble of folds but also on individual folds. This update provides a comprehensive evaluation mechanism that caters to a wider range of use cases.
Commit: ebe4c9de627503b565c18229ea39eb53c8ea10ce
Removal of Unused Argument: The --use_mirroring argument, which was previously set to true in the code regardless of the actual argument passed, has been removed to streamline the code and focus on relevant functionalities.
Commit: 94abee5c2fc13fc0dd5eaf3b74896873c8981194
Goal
The primary goal of this PR is to enhance the flexibility and effectiveness of the inference and evaluation stages of nnU-Net models. By allowing individual folds to be evaluated in addition to the ensemble of folds, users can gain deeper insights into their models' performance and make more informed decisions regarding their deployment.
PR: Enhanced Inference and Evaluation Capabilities for nnU-Net Models
This PR introduces enhancements to the inference and evaluation processes of nnU-Net models, specifically targeting the functionality to execute these processes on individual folds in addition to the existing ensembling approach. Below is a summary of the key changes incorporated into this update:
Changes
Folds Selection for Inference: A new
--folds
argument has been added to therun_inference.py
script, enabling the specification of which folds to use during inference. This feature allows for greater flexibility, as users can now choose to run inference on individual folds or continue using all folds by default, as was previously the case.Commit:
942babf82628c880b9f126234872008e28fa0f6d
Automatic Directory Creation for Evaluation Output: The
run_evaluation.py
script has been updated to automatically create the directory for storing the output.csv
file if it does not already exist. This enhancement simplifies the evaluation process by reducing the need for manual directory management.Commit:
ac8030891948a2a20c5bcde1b8baeae945eefdc7
Enhanced Evaluation Script: Improvements have been made to the
inference_and_evaluation.sh
script to support evaluation not only on the ensemble of folds but also on individual folds. This update provides a comprehensive evaluation mechanism that caters to a wider range of use cases.Commit:
ebe4c9de627503b565c18229ea39eb53c8ea10ce
Removal of Unused Argument: The
--use_mirroring
argument, which was previously set to true in the code regardless of the actual argument passed, has been removed to streamline the code and focus on relevant functionalities.Commit:
94abee5c2fc13fc0dd5eaf3b74896873c8981194
Goal
The primary goal of this PR is to enhance the flexibility and effectiveness of the inference and evaluation stages of nnU-Net models. By allowing individual folds to be evaluated in addition to the ensemble of folds, users can gain deeper insights into their models' performance and make more informed decisions regarding their deployment.