-
Notifications
You must be signed in to change notification settings - Fork 2
Recent Updates
Casper da Costa-Luis edited this page Sep 30, 2024
·
16 revisions
Here you find the recent updates of our challenge:
- Clarify final submission instructions. We ask for up to 3 tags (branches are ignored); consistent with the challenge phase (#115).
- A Torso dataset from a GE Discovery MI 3 ring system is now available on https://petric.tomography.stfc.ac.uk/data/GE_DMI3_Torso/. See also #111. It's TOF data, but restricted to direct sinograms only due to data-sizes etc.
- Siemens mMR NEMA IQ data (low counts) on https://petric.tomography.stfc.ac.uk/data/Siemens_mMR_NEMA_IQ_lowcounts/. This dataset uses the same list-mode data as the other mMR NEMA IQ data, but was unlisted with one fifth of the duration. VOIs are the same. To get reasonable images, the penalisation_factor was changed from the default 1/700 (see the .txt file). See also #108.
- Mediso AnyScan NEMA IQ data now complete and zip file available
- added Mediso AnyScan NEMA IQ data on https://petric.tomography.stfc.ac.uk/data/Mediso_NEMA_IQ/, see SIRF_data_preparation/Mediso_NEMA_IQ and #102.
(
README.md
is currently incomplete, hence no.zip
file yet)
- We have identified a lower precision issue with the calculation of the value of the RDP: it is returned to Python at
float
precision as opposed todouble
. This could affect algorithms which do line searches (when close to convergence). The gradient of the RDP is not affected. Please let us know if this is a problem for you and we can fix the underlying SIRF issue. See #100 for more details.
- added Siemens mMR ACR data on https://petric.tomography.stfc.ac.uk/data/Siemens_mMR_ACR/, see SIRF_data_preparation/Siemens_mMR_ACR and #88.
- other preparation scripts have now been moved into separate directories as well.
- updated Wiki with new (and harder) metric thresholds
- corrected RDP formula on the wiki: add a factor 1/2 (#87)
- reference images and VOIs have been added for the NeuroLF and mMR datasets
-
petric.QualityMetrics
implement the metrics as a callback. These are now added to the leaderboard. -
edge-gpu
image now has a CUDA version of RelativeDifferencePrior (previous version was using the CPU version, which is a lot slower). -
petric.get_data
now returns VOIs, and uses the CUDA RelativeDifferencePrior if it's available.
- The registration for the challenge has opened! Join now!
- There are currently three datasets available on https://petric.tomography.stfc.ac.uk/data. They do not have any region-of-interests nor the reference image yet but these will be added soon. More datasets will follow.
- We have not yet created a final release of SIRF for the challenge. Therefore, if you want to get going, use the
edge
oredge-gpu
Docker build. When building manually, use themaster
branch of the SIRF-SuperBuild. - While we have defined the metrics for evaluation (see the main wiki page), the final thresholds used in the competition will be provided in the following weeks.
- NeuroLF data has been corrected. Headers for all
OSEM_image.hv
are now more complete. -
edge-gpu
image now has the CUDA version of parallelproj (previous version was using the CPU version, which is a lot slower)