Hugging FaceCGANs
Overview
From-scratch Conditional GAN (CGAN) on MNIST. CGANs extend vanilla GANs by conditioning both generator and discriminator on class labels, enabling controlled generation of specific digit classes. Based on Conditional Generative Adversarial Nets (Mirza & Osindero, 2014).
Architecture
Generator: Noise (100D) concatenated with label embedding → ConvTranspose2d layers → 64×64 grayscale image
Discriminator: Image concatenated with label embedding → Conv2d layers → real/fake logit
- InstanceNorm2d, ReLU / LeakyReLU activations
- Weights initialised N(0, 0.02)
- TensorBoard logging
Training
| Hyperparameter | Value |
|---|---|
| Dataset | MNIST (resized to 64×64) |
| Epochs | 30 |
| Batch size | 128 |
| Optimizer | Adam, lr=0.0002, β=(0.5, 0.999) |
| Loss | Binary Cross-Entropy |
Images saved every 500 iterations.
Paper
Conditional Generative Adversarial Nets — Mirza & Osindero, 2014