Issue: Discrepancy Results Between SpikeInterface (using kilosort4 as sorter) and Standalone KiloSort4

Hi, dear spikeinterface developer.

I'm encountering a significant discrepancy between spikeinterface and kilosort 4 when using kilosort 4 example data as dataset. I first preprocess the data using spikeinterface (applying filtering and Common Median Re-referencing (CMR)), and then use spikeinterface.sorters.kilosort4 to sort the preprocessed data. However, the results from this pipeline differ substantially from those obtained by directly running KiloSort4 on the raw example data without any prior preprocessing in SpikeInterface.

Could this difference in results be due to KiloSort4 performing its own internal preprocessing steps, effectively leading to the data being preprocessed twice when using the SpikeInterface pipeline?

And could you please give me some suggestions when using KiloSort4 as a sorter, is it still necessary to perform data preprocessing with SpikeInterface beforehand? If not, what data should be used for calculating extensions when performing quality metrics calculations?

Thanks a lot!

Start by reproducing the comparison on the KiloSort4 example data using SpikeInterface preprocessing with filtering and common median re-referencing, then compare it with standalone KiloSort4 on raw data. Determine how preprocessing affects sorting and which data should be used for quality-metric extensions; done means a documented explanation and recommendation for this workflow.