Table 1 Layer configuration details in the CNN models of the proposed stacked ensemble.
From: Using ensemble learning for classifying artistic styles in traditional Chinese woodcuts
Layer | \(\:\varvec{C}\varvec{N}{\varvec{N}}_{1}\:\left(\varvec{P}\varvec{e}\varvec{r}\varvec{i}\varvec{o}\varvec{d}\right)\) | \(\:\varvec{C}\varvec{N}{\varvec{N}}_{2}\left(\varvec{S}\varvec{t}\varvec{y}\varvec{l}\varvec{e}\right)\) | \(\:\varvec{C}\varvec{N}{\varvec{N}}_{3}\left(\varvec{S}\varvec{t}\varvec{y}\varvec{l}\varvec{e}\right)\) |
|---|---|---|---|
Input | 200 × 200 × 2 | 200 × 200 × 2 | 265 |
Convolution1 | 2D (9 × 9,8) | 2D (10 × 10,8) | 1D (8,16) |
Activation1 | Leaky ReLU | Leaky ReLU | ReLU |
Pooling1 | 2D Avg. pooling (2 × 2) | 2D Avg. pooling (2 × 2) | 1D Max pooling (2) |
Convolution2 | 2D (7 × 7,16) | 2D (8 × 8,16) | 1D (6,24) |
Activation2 | Leaky ReLU | Leaky ReLU | ReLU |
Pooling2 | 2D Avg. pooling (2 × 2) | 2D Avg. pooling (2 × 2) | 1D Max pooling (2) |
Activation3 | Leaky ReLU | ReLU | Sigmoid |
Convolution3 | 2D (5 × 5,24) | 2D (6 × 6,32) | 1D (5,32) |
Pooling3 | 2D Avg. pooling (2 × 2) | 2D Max pooling (2 × 2) | 1D Max pooling (2) |
Convolution4 | 2D (4 × 4,48) | 2D (4 × 4,40) | 1D (4,56) |
Activation4 | ReLU | ReLU | Sigmoid |
Pooling4 | 2D Max pooling (2 × 2) | 2D Max pooling (2 × 2) | 1D Max pooling (2) |
Convolution5 | 2D (3 × 3,56) | 2D (3 × 3,64) | 1D (3,64) |
Activation5 | ReLU | ReLU | Sigmoid |
Pooling5 | 2D Max pooling (2 × 2) | 2D Max pooling (2 × 2) | 1D Max pooling (2) |
FC1 | 80 | 210 | 90 |
FC2 | 3 | 9 | 9 |
