Table 2 Processing benchmarks on synthetic and exemplar images

From: Adaptive particle representation of fluorescence microscopy images

 

Speed up

Time APR (s)

Time pixels (s)

PP

MC pixels (GB)

MC APR (GB)

MRR

Linear neighbor iteration

  CR5

0.55 (0.02)

1.86 (0.07)

1.02 (0.0002)

0.097 (0.003)

3.072 (0)

0.599 (2.7)

5.12 (0.02)

  CR20

1.9 (0.09)

0.54 (0.03)

1.02 (0.002)

0.096 (0.004)

3.072 (0)

0.181 (1.0)

16.9 (0.09)

  CR100

7.1 (0.5)

0.14 (0.009)

1.02 (0.007)

0.076 (0.005)

3.072 (0)

0.053 (0.0005)

60.2 (2.3)

  Exemplars mean

4.06 (5.7)

0.86 (0.5)

1.83 (1.3)

0.094 (0.01)

5.61 (4.2)

0.278 (0.28)

37.5 (56)

Random neighbor access

  CR5

0.71 (0.03)

15.4 (0.2)

11.0 (0.4)

0.126 (0.005)

3.072 (0)

0.599 (2.7)

5.12 (0.02)

  CR20

3.52 (0.3)

3.23 (0.05)

11.4 (0.8)

0.178 (0.01)

3.072 (0)

0.181 (1.0)

16.9 (0.09)

  CR100

24.8 (0.8)

0.44 (0.01)

11.01 (0.3)

0.26 (0.007)

3.072 (0)

0.053 (0.0005)

60.2 (2.3)

  Exemplars mean

11.57 (23.6)

7.29 (10.1)

21.4 (16)

0.17 (0.05)

5.61 (4.2)

0.278 (0.28)

37.5 (56)

Image filtering

  CR5

7.36 (1.2)

1.36 (0.009)

8.02 (0.2)

1.26 (0.2)

4.10 (0)

0.93 (0.003)

4.38 (0.04)

  CR20

14.82 (3.7)

0.76 (0.01)

8.07 (0.3)

0.77 (0.2)

4.10 (0)

0.30 (0.002)

12.82 (0.9)

  CR100

31.10 (13)

0.57 (0.003)

7.96 (0.3)

0.35 (0.15)

4.10 (0)

0.09 (0.0002)

36.85 (6.7)

  Exemplars mean

12.27 (3.0)

1.24 (0.93)

14.13 (9.8)

0.51 (0.33)

7.48 (5.5)

0.36 (0.28)

24.49 (19)

Image segmentation

  CR5

5.10 (0.7)

1.87 (0.02)

8.86 (0.09)

0.86 (0.04)

≈68.5* (0)

12.57 (0.08)

5.51 (0.14)

  CR20

18.30 (2.6)

0.48 (0.003)

8.83 (0.08)

0.95 (0.07)

≈68.5* (0)

3.75 (0.02)

18.18 (0.3)

  CR100

85.3 (12)

0.10 (0.001)

8.78 (0.09)

0.97 (0.06)

≈68.5* (0)

0.80 (0.003)

84.09 (2.9)

  Exemplars mean

N/A

6.99 (5.9)

N/A

N/A

≈385* (286)

13.54 (11.7)

39.72 (40)

  1. For the exemplars, we report the means (standard deviation in brackets) of the values over all exemplar images. For the synthetic fixed-CR datasets, the speed-ups (SU), Pixel-Particle Speed Ratios (PP), and Memory Reduction Ratios (MRR) = (Memory Cost Pixels)/(Memory Cost APR) are averaged over image sizes from 2003 to 10003; absolute timings and Memory Cost (MC) are given for images of size 8003. Graph-cut segmentation on pixels was not possible for 8003 images, as the memory requirement exceeded the 64 GB available on the benchmark machine. The corresponding entries in the table (marked with *) are extrapolated from benchmarks run on smaller images and the SU, PP, and pixel timing for the exemplars could not be determined in this case (N/A). See main text and Supplementary Note 22 for a detailed descriptions of the benchmarks
Back to article page