Fakémon Card Pack Simulator

About the Project

Hi! I'm Oliver, the creator of this website.
In this project I tried to generate Pokémon-like cards using a DCGAN architecture. I hope you have fun trying out the simulator and opening the packs! 😊

🃏 DCGAN Pokémon Card Generator
🎨 Generate traditional-style Pokémon cards using Deep Convolutional Generative Adversarial Networks (DCGANs).

📖 Overview

This project utilizes a DCGAN to create realistic Pokémon cards, focusing on traditional designs from the Pokémon Trading Card Game. The model generates visually convincing cards by training on a dataset of over 11,000 card images spanning all generations (excluding full-art types) to maintain a classic card format.

🚀 Features

• Generate unique, traditional-style Pokémon card designs.
• Trained on a dataset of over 11,000+ Pokémon cards with consistent rectangular dimensions (600x825 pixels).
• Built with PyTorch and trained on an NVIDIA GeForce RTX 3050 (4GB).

🖼️ Dataset

Source: Pokemon TCG Dataset
Size: 11,044 preprocessed images
Preprocessing: Verified for validity, resized to 600x825 pixels, and converted to RGB.

🧠 Model Architecture

Generator: Accepts a 100-dimensional latent vector and outputs a full-color image using 5 transposed convolutional layers, batch normalization, and Tanh activation.
Discriminator: Classifies images as real or fake using 5 convolutional layers with Leaky ReLU activations and a Sigmoid output.

📊 Results

The training losses of the Generator and Discriminator as well as a Turing test comparing real vs. generated cards are documented below.