GAN Architecture
Generator + discriminator adversarial loop with loss flow.
Rendering…
Make it your own.
flowchart LR
Z["Noise z ~ N(0,1)"] --> G[Generator]
G --> Fake[Fake sample]
Real[Real sample] --> D[Discriminator]
Fake --> D
D --> Pred{Real or fake?}
Pred --> Ldis["Discriminator loss<br/>-log D(real) - log(1-D(fake))"]
Pred --> Lgen["Generator loss<br/>-log D(fake)"]
Ldis -.->|gradient| D
Lgen -.->|gradient| G
classDef adv fill:#fee2e2,stroke:#991b1b;
class G,D adv