Synapse tfrecords normalization

[anoukstein]

Thanks for including your model weights! I'm able to see predictions on the Synapse data but not on tfrecords created from npy.h5 created from DICOM images. I notice that the Synapse data is not normalized to between 0 and 1 but instead I'm getting the following values for max/min. Can you comment on the normalization step?
1.1594521 -0.10392588 float32
1.1950973 -0.10913128 float32
1.0950696 -0.0893049 float32
1.1503692 -0.097785614 float32
1.1254649 -0.10286699 float32
1.1124429 -0.1018485 float32
1.1358986 -0.15249471 float32

[kenza-bouzid]

Hi Anouk, I suggest that you directly reach out to the authors of the original paper. I have requested the preprocessed data as suggested in the repo https://github.com/Beckschen/TransUNet cause we were running out of time for the uni project. Jieneng Chen provided us with the processed data in npy and h5 format. My team and I tried to reproduce the paper with the exact same settings, however we encountered some irregularities especially while handling the data. We also noticed that the min/max values didn't comply with the normalization step claimed by the original authors. We decided to proceed with the provided data as our goal was to reproduce the paper. You can however still get the raw data from https://www.synapse.org/#!Synapse:syn3193805/wiki/ and process it from there.

…

On Thu, 12 Aug 2021 at 18:45, Anouk Stein ***@***.***> wrote: Thanks for including your model weights! I'm able to see predictions on the Synapse data but not on tfrecords created from npy.h5 created from DICOM images. I notice that the Synapse data is not normalized to between 0 and 1 but instead I'm getting the following values for max/min. Can you comment on the normalization step? 1.1594521 -0.10392588 float32 1.1950973 -0.10913128 float32 1.0950696 -0.0893049 float32 1.1503692 -0.097785614 float32 1.1254649 -0.10286699 float32 1.1124429 -0.1018485 float32 1.1358986 -0.15249471 float32 — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#5>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AI5J63HMVKDRDDOBHU2MYQTT4P3DLANCNFSM5CBU34HQ> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&utm_campaign=notification-email> .

[anoukstein]

Ah, that makes sense. That helps me that you noticed the same min/max discrepancies. I will reach out to them. Nice job on this by the way, hope you got a good grade!!