unetr_plus_fusion_hfs: A Python repository from Allen-lz

bug1:

ValueError: Invalid type <class 'numpy.int32'> for the flop count! Please use a wider type to avoid overflow

https://github.com/facebookresearch/fvcore/issues/104

eval synapse

验证的时候需要对脚本上的路径进行更改之后才能正常使用

#!/bin/sh

DATASET_PATH=D:/datasets/UNETR_PP_DATASETS/DATASET
CHECKPOINT_PATH=unetr_pp/evaluation/unetr_pp_synapse_checkpoint

export PYTHONPATH=.././
export RESULTS_FOLDER="$CHECKPOINT_PATH"
export unetr_pp_preprocessed="$DATASET_PATH"/unetr_pp_raw/unetr_pp_raw_data/Task02_Synapse
export unetr_pp_raw_data_base="$DATASET_PATH"/unetr_pp_raw

python unetr_pp/run/run_training.py 3d_fullres unetr_pp_trainer_synapse 2 0 -val

并且在run_training.py上需要增加路径

import sys
sys.path.append("..")
sys.path.append("../..")
sys.path.append(".")

训练的时候保存的配置与模型也要放到相应的路径上

unetr_pp/evaluation/unetr_pp_synapse_checkpoint/unetr_pp/3d_fullres/Task002_Synapse/unetr_pp_trainer_synapse__unetr_pp_Plansv2.1/fold_0/model_final_checkpoint.model
unetr_pp/evaluation/unetr_pp_synapse_checkpoint/unetr_pp/3d_fullres/Task002_Synapse/unetr_pp_trainer_synapse__unetr_pp_Plansv2.1/fold_0/validation_raw/validation_args.json

输入以下命令才能正常的验证(注意更换数据集的路径), 这个脚本是生成一些结果并保存下来

bash evaluation_scripts/run_evaluation_synapse.sh

要得到真正的指标需要使用到推理

python unetr_pp/inference_synapse.py fold_0

Allen-lz/unetr_plus_fusion_hfs

bug1:

eval synapse