preprocessing

Hi,

I have a question regarding preprocessed data. You said that "Note that private test data is also preprocessed so we recommend the use of the preprocessed set". I am a little bit confused here, do you mean that input for the test submissions will be the preprocessed data?

Thanks in advance, Bests ZP

Hi Zeynep,

Yes, the input for the test submissions will be preprocessed in the same way as it was done in the provided preprocessed data. So, we will apply preprocessing on all the test CXR images.

Best, Ecem

Thanks for your quick reply. In that case, a model trained with original data + ad-hoc preprocessing won't fit at all to test dataset.

Bests ZP

Hi Zeynep,

I think the model trained with original data would still perform well on the test data. We have performed this type of experiments, we do not expect to see much difference on performance when the model is trained with original or preprocessed dataset.

We wanted to standarize the appearance of the test images so that different preprocessing methods would not have a huge performance impact on the final performance, but we could judge the "models" better. For that reason we have decided to provide preprocessed data and to make the code for this preprocessing publicly available. This levels the playing field for participants.

Hope this clarifies things a bit, please let us know if you have any question.

Best, Ecem