GAN-MNIST

A simple yet effective DCGAN trained on the MNIST dataset using PyTorch, designed for data augmentation experiments.

🧠 Model Details

• Architecture: Deep Convolutional GAN (DCGAN)
• Generator/Discriminator: 3-layer CNN
• Latent Dimension: 100
• Epochs Trained: 20
• Final FID Score: 24.16
• Image Size: 28×28 grayscale

📦 Usage

from torch import load
from models.gan.model import Generator

# Load model
model = Generator(latent_dim=100, img_channels=1)
model.load_state_dict(load("generator.pt"))
model.eval()