search

hubmapconsortium / ingest-pipeline

Data ingest pipeline(s) for QA/metadata etl/post-processing
MIT License
4 stars 4 forks source link

MALDI Pyramids Look Odd at Low Resolution Levels #108

Closed ilan-gold closed 3 years ago

ilan-gold commented 4 years ago

If you look at the lower resolution levels, it looks like something went wrong with the downsampling, where there are these stripes:

Screen Shot 2020-07-18 at 6 50 37 PM Screen Shot 2020-07-18 at 6 50 09 PM

This could be an issue with the data being 32 bit float (I still don't understand why it has to be this data type) or this could be something else. Perhaps the pyramid generator relying just on dropout and not a gaussian filter + dropout is biting us. I have no insight on to this one - if it is a real issue, and we can reproduce it, I would just report it to glencoe.

I have confirmed this using Fiji (i.e not just our image viewer).

Here are the UUIDs:

PROD: 146063814f6bd2b6546e950d0ca5f8c3,HBM645.ZQSN.258,['MALDI-IMS-neg'],Vanderbilt TMC
TEST: 71f6fe348843c65e3af47388d181c0e6,HBM587.HGDH.925,['MALDI-IMS-neg'],IEC Testing
ilan-gold commented 4 years ago

One option is to just not run the pyramid pipeline on the IMS images.

icaoberg commented 4 years ago

@ilan-gold recreated results so you can take a look at the logs on TEST IEC Testing Group/097358f1cb26c275adbeabfe4ceca8a9/