Generator Architecture
Defines a generator using Dense, Conv2DTranspose, and BatchNorm layers. Assumes upsampling via stride=2 ConvTranspose layers.
Initial Dense & Reshape
Initial Dense output units = Start H * Start W * Start Channels.
Upsampling Block 1 (ConvT + BN + ReLU)
Upsampling Block 2 (ConvT + BN + ReLU)
Final Layer (ConvT + Tanh)
Outputs the final image.
Discriminator Architecture
Defines a discriminator using Conv2D, Dense, and Embedding layers. Assumes downsampling via stride=2 Conv2D layers. Outputs predictions for image source (real/fake) and class label.
Downsampling Block 1 (Conv2D + LeakyReLU)
Downsampling Block 2 (Conv2D + LeakyReLU)
Flatten & Final Dense Layer
Image features are flattened after Conv blocks. This dense layer processes the flattened features before the output heads.
Estimated Parameters
Click 'Calculate Parameters' to see the estimated counts based on your defined architecture.
Parameter Counts
Generator Parameters:
Discriminator (Source Path):
Discriminator (Class Path):
Discriminator (Total):
Total Estimated Parameters:
