Skip to main content

Table 4 Average processing time of basic operations in experimented models

From: Estimation of gait normality index based on point clouds through deep auto-encoder

Model

Framework

Preprocessing (using C++)

Forward and backward (in training stage)

Forward (in inference stage)

FoldingNet [39]

TensorFlow

0.262 (ms)

1.639 (s)

0.446 (s)

PointNet [24]

TensorFlow

0.262 (ms)

1.308 (s)

0.102 (s)

RSNet [14]

Torch

0.311 (ms)

0.202 (s)

0.058 (s)

Our 6 models

TensorFlow

1.126 (ms)

0.014 (s)

0.002 (s)

  1. The preprocessing indicates the cylindrical histogram formation in our method and the cloud downsampling in the others. The time is reported in seconds and milliseconds
  2. The italic values correspond to fastest running speeds in execution stages