You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In the testing script located at "examples/advanced/brats18/result_stat/brats_3d_test_only.py"
The script uses site-All.json to locate test cases. However, some cases in site-All.json overlap with site-4.json, which was already used during training. Testing on previously seen data inflates performance metrics. So, the reported test results do not accurately measure the model's ability to segment truly unseen data.
I think modifying site-All.json to exclude cases used before, or use a separate dataset for testing that was never seen during training or validation will provide true result. Thank you
The text was updated successfully, but these errors were encountered:
In the testing script located at "examples/advanced/brats18/result_stat/brats_3d_test_only.py"
The script uses site-All.json to locate test cases. However, some cases in site-All.json overlap with site-4.json, which was already used during training. Testing on previously seen data inflates performance metrics. So, the reported test results do not accurately measure the model's ability to segment truly unseen data.
I think modifying site-All.json to exclude cases used before, or use a separate dataset for testing that was never seen during training or validation will provide true result. Thank you
The text was updated successfully, but these errors were encountered: