dcgan

Deep Convolutional GAN learns to generate MNIST digits

Real Images

Start training to see real samples

Generated Images

Generated samples will appear here
Discriminator Loss
-----StepLoss
-
Generator Loss
-----StepLoss
-
Idle
Epoch 0/100Batch 0-
loss
-

Hyperparameters

0.0002
100
100

About DCGAN

Deep Convolutional GANs use Conv2d layers in the discriminator and ConvTranspose2d layers in the generator for upsampling. The generator learns to produce realistic images from random noise, while the discriminator learns to distinguish real from fake.

This architecture uses strided convolutions instead of pooling layers, and batch normalization for stable training.

Logs
No output yet...