Display Settings:

Items per page
We are sorry, but NCBI web applications do not support your browser and may not function properly. More information

Results: 7

1.
Figure 1

Figure 1. From: CUDA Optimization Strategies for Compute- and Memory-Bound Neuroimaging Algorithms.

CUDA computational times of computing the force field with varying volume dimensions using the recompute and reuse methods with different thread block sizes.

Daren Lee, et al. Comput Methods Programs Biomed. ;106(3):175-187.
2.
Figure 2

Figure 2. From: CUDA Optimization Strategies for Compute- and Memory-Bound Neuroimaging Algorithms.

Performance metrics of the flow field kernel with varying volume dimensions of (a) the speedup of the optimized GTX 480 reuse method versus the CPU one, and (b) the giga-flops (GFLOPS) for the GTX 480 implementation.

Daren Lee, et al. Comput Methods Programs Biomed. ;106(3):175-187.
3.
Figure 6

Figure 6. From: CUDA Optimization Strategies for Compute- and Memory-Bound Neuroimaging Algorithms.

Results of applying the level set based non-local surface denoising technique to an artificially created noisy surface of two square blocks using the CPU and GPU implementations. The maximum error for the colorbar is 0.5 voxels.

Daren Lee, et al. Comput Methods Programs Biomed. ;106(3):175-187.
4.
Figure 4

Figure 4. From: CUDA Optimization Strategies for Compute- and Memory-Bound Neuroimaging Algorithms.

CUDA performance times to compute the patch weights in the non-local surface denoising algorithm with varying narrow band size and with different methods to store the subset of patches compared during each pass using global, texture, and constant memory.

Daren Lee, et al. Comput Methods Programs Biomed. ;106(3):175-187.
5.
Figure 5

Figure 5. From: CUDA Optimization Strategies for Compute- and Memory-Bound Neuroimaging Algorithms.

Performance metrics of the non-local surface denoising algorithm with varying narrow band size of (a) the overall speedup of the optimized GTX 480 implementation versus the dual-thread CPU one and (b) the gigabytes per seconds (GB/s) for the GTX 480 implementation.

Daren Lee, et al. Comput Methods Programs Biomed. ;106(3):175-187.
6.
Figure 7

Figure 7. From: CUDA Optimization Strategies for Compute- and Memory-Bound Neuroimaging Algorithms.

Results of applying the GPU implementation of the level set based non-local surface denoising to white matter surfaces segmented from human subjects. The original surfaces are on the left and the smoothed surfaces are on the right.

Daren Lee, et al. Comput Methods Programs Biomed. ;106(3):175-187.
7.
Figure 3

Figure 3. From: CUDA Optimization Strategies for Compute- and Memory-Bound Neuroimaging Algorithms.

Results of applying the GPU implementation of the 3D unbiased image registration to 192 × 128 × 192 human MRI cases aligned to a template volume. The middle sagittal slice for each of the volumes are shown. The original image volumes are in the first column, the template volume is in the second column, and the aligned output images overlaid with the deformation vector field are shown in the third and fourth columns for the CPU and GPU. The last columns shows the difference between the vector magnitudes in units of voxels. The maximum value for the colorbar is 0.5 voxels.

Daren Lee, et al. Comput Methods Programs Biomed. ;106(3):175-187.

Display Settings:

Items per page

Supplemental Content

Recent activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...
Write to the Help Desk