Table 2 Discriminator network architecture: layer types, activations, output shapes (channels × height × width) and number of trainable parameters for each layer. All convolutional layers have stride =2. LeakyReLU’s leakines =0.2

From: CosmoGAN: creating high-fidelity weak lensing convergence maps using Generative Adversarial Networks

	Activ.	Output shape	Params.
Input map	–	1 × 256 × 256	–
Conv 5 × 5	LReLU	64 × 128 × 128	1664
Conv 5 × 5	–	128 × 64 × 64	205K
BatchNorm	LReLU	128 × 64 × 64	256
Conv 5 × 5	–	256 × 32 × 32	819K
BatchNorm	LReLU	256 × 32 × 32	512
Conv 5 × 5	–	512 × 16 × 16	3.3M
BatchNorm	LReLU	512 × 16 × 16	1024
Linear	Sigmoid	1	131K
Total trainable parameters			4.4M

Back to article page