Table 5 Execution times of training and inference on GPU and CPU.

From: Investigation of an efficient multi-modal convolutional neural network for multiple sclerosis lesion detection

Approach

Inference GPU

Inference CPU

Train one fold

GFLOPs

RAM

VRAM

2.5D Tiramisu 17

7 s

5.5min

12.5 h

23.9

6 GB

19.3 GB

nnU-Net19

15 s (5 s)

207 min (5.2 min)

34.7 h (10.4 h)

2.74 e+03

22 GB

9.2 GB

nnU-Net small20

12 s (4 s)

164 min (4.2 min)

29.4 h (9.2 h)

2.21 e+03

22 GB

8.9 GB

Multi-branch U-Net (proposed)

4 s

67 s

4.2 h

53.2

4GB

9.4 GB

  1. This table compares the duration times of two other publicly available state-of-the-art architectures on our systems, as well as the performance metrics from the ISBI challenge submission. The best and second-best results are written in bold and italic, respectively.
Back to article page