Difference between revisions of "Slicer-IGT/GPU-IGT/112707"
From NAMIC Wiki
| Line 10: | Line 10: | ||
Rigid registration using affine transformation, MI similarity measure, and Powell optimization method impletented on CUDA platform. GPU-acceleration was performed on affine transformation and MI similarity measure. Performance boost of around x12 was achieved. | Rigid registration using affine transformation, MI similarity measure, and Powell optimization method impletented on CUDA platform. GPU-acceleration was performed on affine transformation and MI similarity measure. Performance boost of around x12 was achieved. | ||
| − | * Non-rigid registration | + | * Non-rigid registration (also Japanese cas paper) |
| + | [[Image:NonrigidReg.jpg]] | ||
| + | |||
| + | Non-rigid registration using Rueckert's B-spline algorithm. Performance boost of around x10-x20 was achieved. | ||
#Tool kit used | #Tool kit used | ||
*CUDA calculator | *CUDA calculator | ||
| − | *8800GTX is CUDA 1.0 compatible | + | *8800GTX, 8800GTS, Quadro FX5600, Tesla C870 is CUDA 1.0 compatible |
| − | * | + | *8600GTS, 8800GT is CUDA 1.1 compatible which comes with Atomic function to control cuncurrent access to memoery from multiple thread |
# Extension to ITK | # Extension to ITK | ||
Revision as of 14:23, 6 December 2007
Home < Slicer-IGT < GPU-IGT < 112707- Introduction of members of this project
- Nicholas: student from Tokyo
- Benjamin: student from ETH, be at SPL for 6 mo. till May 31, 2007. Experience in golfing simulator, histoscopy simulator, open-source game software. His project here is GPU accelerated Slicer for 4D IGT.
- Nicholas update on his CUDA project
- Rigid and non-rigid registration using CUDA using Nvidia 8800 GTX (350GFlps) compatible with CUDA platform.
- Rigid registration (Japanese cas paper)
Rigid registration using affine transformation, MI similarity measure, and Powell optimization method impletented on CUDA platform. GPU-acceleration was performed on affine transformation and MI similarity measure. Performance boost of around x12 was achieved.
- Non-rigid registration (also Japanese cas paper)
Non-rigid registration using Rueckert's B-spline algorithm. Performance boost of around x10-x20 was achieved.
- Tool kit used
- CUDA calculator
- 8800GTX, 8800GTS, Quadro FX5600, Tesla C870 is CUDA 1.0 compatible
- 8600GTS, 8800GT is CUDA 1.1 compatible which comes with Atomic function to control cuncurrent access to memoery from multiple thread
- Extension to ITK
- CMake turn on/off #DEFINE
- VTK VolumePro as part of volume redering
- ITK parallelized process
- [Action items, Nichoals] Mid-term goal for Nicholas is to port his rigid and non-rigid regstration to ITK
- [Action items, Nicholas] Contact Utah team hear how exactly they implement their ITK.
- [Action items, Nicholas] sending volume rendering code by Dec. 6th.
- Volume rendering
- CUDA accelerated volume rendering
- x15 - 20 improvement
- comparison to other people's CG-based volume rendering media:SIGGRAPH-GPU.pdf
- [Action items, Benjamin] port CUDA-based volume rendering to vtk volume rendering classes, and then to Slicer
- [Action items, Benjamin] succeed Nicholas' itk-cuda-rigid non-rigid regstration and port them to Slicer in the context of MRg cardiac ablation
- Timeline
- NH will write paper on IV-CUDA rendering for Journal
- Hata suggested publication of ITK-CUDA registration in Insight Journal (Feb)
- Benjamin will have short summery as of March 15 for MICCAI.
- Benjamin will finish his project by May 15th.
- Benjamin
- Communication
Bi-weekly t-con to update each other week. Next one is Dec 11, 2007 at 9am.
