Cannot be inferred using the last model provided？

[liaochuanlin]

Task003_tumor Cannot be inferred using the last model provided？
Traceback (most recent call last):
File "/home/lcl/unetr_plus_plus/unetr_pp/training/model_restore.py", line 168, in
trainer = restore_model(pkl, checkpoint, train)
File "/home/lcl/unetr_plus_plus/unetr_pp/training/model_restore.py", line 58, in restore_model
info = load_pickle(pkl_file)
File "/home/lcl/anaconda3/envs/unetr_pp/lib/python3.8/site-packages/batchgenerators/utilities/file_and_folder_operations.py", line 50, in load_pickle
a = pickle.load(f)
_pickle.UnpicklingError: A load persistent id instruction was encountered,
but no persistent_load function was specified.

[Amshaker]

Could you please share the command you using for evaluation?

[auroua]

@Amshaker

I encounter the same problem, and find that the evaluation code of BraTS is different from the other datasets.

The run_evaluation_turom.sh uses the unetr_pp/inference/predict_simple.py to generate the prediction, while other datasets using the run_training.py to generate segment masks.

The predict_simple.py calls the function restore_model from unetr_plus_plus/unetr_pp/training/model_restore.py to load the pre-trained weights.

The first line of restore_model loads the pre-trained weights in the .pkl format, but the pre-trained weights are in the .model format, which causes the problem above.

You can repeat the above error by running the evaluation_scripts/run_evaluation_turmor.sh directly.

Could you provide the pre-trained weights in .pkl format, as the mix of training and inference code together is hard to use.

I think inference the segment mask of raw input using the predict_simple.py is more reasonable than using the unetr_pp/run/run_training.py.