Table 8 Inference efficiency comparison on an NVIDIA RTX A6000 GPU (FP16 precision)

From: Structure-aware multi-task learning with domain generalization for robust vertebrae analysis in spinal CT

Method

Vol/s

Latency/Volume (ms)

nnU-Net7

9.4

106.4

UNETR19

8.3

120.5

TransBTS46

10.2

98.0

H2Former22

7.8

128.2

Scribformer23

8.7

114.9

Dense-U-Net6

6.3

157.8

Tao et al.49

7.1

140.8

VertebraFormer (ours)

13.8

72.5

  1. Throughput is measured in 3D volumes processed per second (Vol/s) for 1283 inputs, including all decoding and post-processing41.
Back to article page