Closed anhtuduong closed 1 week ago
@anhtuduong I sent the request for full test data to the owner Nicholas Heller 3 days ago but he hasn't replied. If the situation remains unchanged to the end of this week, maybe we can try make our own test cases by cut 10% of current dataset as you offered. I will update asap !
@thuhoainguyen I see. In this case, I will prepare a script to split the test set. Hope that we don't need to use it...
@anhtuduong good news ! He sent me the test data today !! Here is the link to download: https://kits23-test-data.s3.us-east-2.amazonaws.com/230720_kits23_test_images.tar?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Credential=AKIA6LZQ3U2PKOLZDEXO%2F20240616%2Fus-east-2%2Fs3%2Faws4_request&X-Amz-Date=20240616T144324Z&X-Amz-Expires=604800&X-Amz-SignedHeaders=host&X-Amz-Signature=190acdc9c59ab88436ffaa3db594ed3d46305a41b4b5e6dd0b702e82d0588bc9
@thuhoainguyen The dataset is composed of
599 cases
with489 allocated to the training set
and110 in the test set
. However these110 test set
were not published. There are 2 scenarios we are facing now:Obtaning the test cases: https://discourse.kits-challenge.org/t/how-to-obtain-test-data-in-the-kits23-dataset/506
Make our own test cases: If cannot reach the official test cases, we have to cut 10% of our dataset for the test cases. Which mean from the
original 489 cases
, we divide into:440 for training
and49 for test
.Having test set is crucial because we need to evaluate the model after the training. I suggest you try to contact the guy in the first link, or ask Jacopo about this. In the mean time I will setup the format for nnUNet 😎