Question about test set

Will the private test set be a "new Study" (so another microscope (or different model and parameters), another sample species, a different voxel size, ...) or will all the samples in the private test set match at least one sample from the training set?

The “test phase” private set contains the same image studies as the training database, but with different time steps. We've chosen to do this because this phase is primarily intended to help you create your dockers containers.

The private set of the “evaluation phase” will contain further image studies identical to those in the train database, as well as new, previously unpublished images (but not too far from the train images).

Hi everyone,

given that the test set may contain new studies, it's probably not advisable to use a study-specific model, but rather a generalizing model if possible? Or will the study ids of the test set also be set such that they match the specimen / setup of the training studies?

Thanks and best regards,

Johannes

Hi,

Thanks for that question. You are totally free to choose the model you want to use ! ;)

I've just updated the metadata for all images for the Test and the Evaluation phases. Then you can access to the channel, study, PhysicalSizeX, PhysicalSizeY, PhysicalSizeZ, axes and shape metadata - like the images in the Training database.

As a reminder, there are no new image studies in the test phase, even if the images are not identical to those in the training database.

For the evaluation phase, the test set will contain 1/4 new image studies (not too different) and 3/4 similar image studies.

Best,

Dorian