Age and Gender Prediction using Deep cnns and Transfer Learning

Download 299,02 Kb. Pdf ko'rish
bet	5/13
Sana	22.12.2023
Hajmi	299,02 Kb.
	#127116

1 2 3 4 5 6 7 8 9 ... 13

Bog'liq
xudo xoxlasa tushadi99%, 3-labarotoriya ishi Saralash usul va algoritmlarini tadqiq qilis, cmd buyruqlari, Incremental model nima, 1matematik, word sAM 1 savol, Документ Microsoft Word (4), Ma\'ruzalar (2), ЛАБОРАТОРНАЯ РАБОТА N1, Dasturlash 2, Ariza, Qalandarova Gulshoda, 1648631455, 1650692784, 1651669892 (2)

3.2
Deep CNNs
Network Architecture. The tasks tackled using the deep CNN approach include age
and gender classification and age estimation. The basic structure of each of the 3 models
includes a series of convolutional blocks, followed by a set of FC (fully connected)
layers for classification and regression. An RGB image is fed to the model and is resized
to 180 x 180 x 3. Every architecture comprises convolutional blocks that are a stack of
convolutional layers (filter size is 3x3) followed by non-linear activation ‘ReLU’, max
pooling (2x2) and batch normalization to mitigate the problem of covariate shift. The
deeper layers here also have spatial dropout (drop value of 0.15-0.2)
which drops entire
feature maps to promote independence between them. Following the convolutional
blocks, the output is flattened before feeding that into FC layers. These FC layers have
activation function of ReLU, dropout (value between 0.2 & 0.4) and batch normaliza-
tion. Table 3 shows the architecture used for age estimation.
The architectures for age classification and gender classification differ in the fact
that they have 3 & 2 blocks with 256 filters respectively (in convolutional layer) and
the output layer has 5 and 2 neurons respectively with softmax activation function (be-
ing classification tasks).

5

Table 3. Network Architecture for Age Estimation
Layer
Filters
Output Size
Kernel Size
Activation
Image
-
180 x 180 x 3
-
-
Separable Conv1
64
180 x 180 x 64
3 x 3
ReLU
Max Pooling
-
90 x 90 x 64
2 x 2
-
Separable Conv2
128
90 x 90 x 128
3 x 3
ReLU
Max Pooling
-
45 x 45 x 128
2 x 2
-
Separable Conv3
128
45 x 45 x 128
3 x 3
ReLU
Max Pooling
-
22 x 22 x 128
2 x 2
-
Separable Conv4
256
22 x 22 x 256
3 x 3
ReLU
Max Pooling
-
11 x 11 x 256
2 x 2
-
Separable Conv5
256
11 x 11 x 256
3 x 3
ReLU
Max Pooling
-
5 x 5 x 256
2 x 2
-
FC1
-
128
-
ReLU
FC2
-
64
-
ReLU
FC3
-
32
-
ReLU
Output
-
1
-
ReLU

Download 299,02 Kb.

1 2 3 4 5 6 7 8 9 ... 13

Download 299,02 Kb.

Pdf ko'rish