Data anomaly report

I create this topic to report inconsistency problems I found in the data. If you found anomaly, irregularity or issue, this may be the place to share them. To try to make this topic clearer, I suggest reserving a message for each issue.

Starting from image 213 (I checked up to 480 right now) the metadata in the angle image is missing study and channel metadata.

This is not a big issue as the fused image still have it.

EDIT : the message was updated when I increased the image I checked.

I have a quick question on the percentile normalization that will be applied before the comparison of predicted and ground truth images: it seems the implementation you're using is not clipping the range after the normalization, i.e., it could happen that there will be values larger than 1 after normalization. Is this intended or would it make sense to clip the normalized result to the range [0,1]?

Thanks in advance and best regards,

Johannes

In photonic microscopy, some pixel values can be very strong (outliers), well beyond the standard values obtained during acquistion. Percentile normalization makes it possible to normalize an image to the mean distribution of its pixels, without these outliers disturbing the image. These few outliers appear after normalization in values <0 or >1.

Thanks for the reply – jop sure that totally makes sense. I used to clip those values to the range 0-1 after the normalization, but this then artificially moves the outliers to within the bounds.