Fine-Tuned Temporal Dense Sampling with 1D Convolutional Neural Network for Human Action Recognition

. 2023 Jun 2;23(11):5276. doi: 10.3390/s23115276

Algorithm 1 Training in the proposed FTDS-1DConvNet Framework

1:
Initialization: All parameters and hyperparameters.
2:
Fine-tuning: Pre-trained Inception-ResNet-V2
3:
for $e p o c h = 1$ to $e p o c h = M a x i m u m E p o c h$ do
4:
for $i t e r = 1$ to $i t e r = M a x i m u m I t e r a t i o n$ do
5:
$𝒜 \leftarrow$ FTDS (refer to (6))
6:
Hidden states calculation: $ℋ^{l}$ (refer to (8))
7:
Output layer calculation: $\vec{z}$ (refer to (9))
8:
Probability calculation: $\vec{\hat{y}}$ (refer to (11))
9:
Concatenation of the parameters: $θ = (W, \vec{b})$
10:
Gradients calculation: $\frac{α}{\sqrt{{\hat{v}}_{t}} + ϵ} * {\hat{m}}_{t}$ (refer to (14) and (15))
11:
Parameters update: $θ \leftarrow θ - \frac{α}{\sqrt{{\hat{v}}_{t}} + ϵ} * {\hat{m}}_{t}$
12:
end for
13:
end for