rmachado23/pokemon-gan

Pokemon DCGAN Generator

Unconditional DCGAN trained on Pokemon sprites to generate 64x64 RGB images.

Architecture

• Type: DCGAN Generator
• Input: noise vector (100,)
• Output: RGB image (3, 64, 64), Tanh activation [-1, 1]
• Parameters: ~3.5M

Training

• Dataset: huggan/pokemon (~800 images)
• Epochs: 200
• Optimizer: Adam (lr=0.0002, betas=(0.5, 0.999))
• Loss: BCELoss with label smoothing (0.9)
• G steps per D step: 2

Usage

from pokemon_gan_model import Generator
import torch

G = Generator()
G.load_state_dict(torch.load("pokemon_gan_generator.pth", map_location="cpu"))
G.eval()

z = torch.randn(1, 100)
with torch.no_grad():
    img = G(z)  # (1, 3, 64, 64) in [-1, 1]