Architecture for our autoencoder, featuring both encoder and decoder subnets. The decoder is a direct reflection of the encoder’s structure.
| . | Layer . | Layer type . | Units/number of filters . | Size . | Padding . | Stride . |
|---|---|---|---|---|---|---|
| Encoder | Input | Input | – | (64,64) | – | – |
| Conv1 | Convolutional | 8 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Conv2 | Convolutional | 8 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| MaxPool | Max-pooling | – | (2,2) | – | 1 | |
| Conv3 | Convolutional | 16 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Conv4 | Convolutional | 16 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| MaxPool | Max-pooling | – | (2,2) | – | 1 | |
| Linear | Fully-connected | 3 | – | – | – | |
| Decoder | Linear | Fully-connected | 3 | – | – | – |
| Up | Partial inverse max-pool | – | (2,2) | – | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans1 | Transposed Convolution | 16 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans2 | Transposed Convolution | 16 | (3,3) | 1 | 1 | |
| Up | Partial inverse max-pool | – | (2,2) | – | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans3 | Transposed Convolution | 8 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans4 | Transposed Convolution | 8 | (3,3) | 1 | 1 | |
| Ouput | Output | – | (64,64) | – | – |
| . | Layer . | Layer type . | Units/number of filters . | Size . | Padding . | Stride . |
|---|---|---|---|---|---|---|
| Encoder | Input | Input | – | (64,64) | – | – |
| Conv1 | Convolutional | 8 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Conv2 | Convolutional | 8 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| MaxPool | Max-pooling | – | (2,2) | – | 1 | |
| Conv3 | Convolutional | 16 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Conv4 | Convolutional | 16 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| MaxPool | Max-pooling | – | (2,2) | – | 1 | |
| Linear | Fully-connected | 3 | – | – | – | |
| Decoder | Linear | Fully-connected | 3 | – | – | – |
| Up | Partial inverse max-pool | – | (2,2) | – | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans1 | Transposed Convolution | 16 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans2 | Transposed Convolution | 16 | (3,3) | 1 | 1 | |
| Up | Partial inverse max-pool | – | (2,2) | – | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans3 | Transposed Convolution | 8 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans4 | Transposed Convolution | 8 | (3,3) | 1 | 1 | |
| Ouput | Output | – | (64,64) | – | – |
Architecture for our autoencoder, featuring both encoder and decoder subnets. The decoder is a direct reflection of the encoder’s structure.
| . | Layer . | Layer type . | Units/number of filters . | Size . | Padding . | Stride . |
|---|---|---|---|---|---|---|
| Encoder | Input | Input | – | (64,64) | – | – |
| Conv1 | Convolutional | 8 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Conv2 | Convolutional | 8 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| MaxPool | Max-pooling | – | (2,2) | – | 1 | |
| Conv3 | Convolutional | 16 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Conv4 | Convolutional | 16 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| MaxPool | Max-pooling | – | (2,2) | – | 1 | |
| Linear | Fully-connected | 3 | – | – | – | |
| Decoder | Linear | Fully-connected | 3 | – | – | – |
| Up | Partial inverse max-pool | – | (2,2) | – | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans1 | Transposed Convolution | 16 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans2 | Transposed Convolution | 16 | (3,3) | 1 | 1 | |
| Up | Partial inverse max-pool | – | (2,2) | – | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans3 | Transposed Convolution | 8 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans4 | Transposed Convolution | 8 | (3,3) | 1 | 1 | |
| Ouput | Output | – | (64,64) | – | – |
| . | Layer . | Layer type . | Units/number of filters . | Size . | Padding . | Stride . |
|---|---|---|---|---|---|---|
| Encoder | Input | Input | – | (64,64) | – | – |
| Conv1 | Convolutional | 8 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Conv2 | Convolutional | 8 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| MaxPool | Max-pooling | – | (2,2) | – | 1 | |
| Conv3 | Convolutional | 16 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Conv4 | Convolutional | 16 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| MaxPool | Max-pooling | – | (2,2) | – | 1 | |
| Linear | Fully-connected | 3 | – | – | – | |
| Decoder | Linear | Fully-connected | 3 | – | – | – |
| Up | Partial inverse max-pool | – | (2,2) | – | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans1 | Transposed Convolution | 16 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans2 | Transposed Convolution | 16 | (3,3) | 1 | 1 | |
| Up | Partial inverse max-pool | – | (2,2) | – | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans3 | Transposed Convolution | 8 | (3,3) | 1 | 1 | |
| ReLU | Activation | – | – | – | – | |
| Trans4 | Transposed Convolution | 8 | (3,3) | 1 | 1 | |
| Ouput | Output | – | (64,64) | – | – |
This PDF is available to Subscribers Only
View Article Abstract & Purchase OptionsFor full access to this pdf, sign in to an existing account, or purchase an annual subscription.
