🎨 Image generator - GAN

📝 Project Description

This repository implements a Generative Adversarial Network (GAN) using PyTorch.
It allows training and using a generator model to produce realistic images of Cat🐈 from random noise vectors.
The project is designed to be modular, so you can train your own GAN on a custom dataset or load pre-trained models to generate images quickly.

For me, I used a dataset of cat pictures to train it ! 😺

---

⚙️ Features

• 🎨 Train a GAN on your own dataset of images.
• 💾 Save and reload both generator and discriminator models.
• 🎲 Generate batches of fake images directly from random noise.
• 📊 Visualization of generated samples during training.

🖼️ Example of generated output :

🔍 Preliminary Results:
Some generated images show recognizable cat features ✅, but results still need improvement.
Longer training (~13h GPU 💻🔥) would be required for sharper, more realistic outputs.

---

⚙️ How It Works

• 🎨 Generator (GNet) Transforms a random noise vector (latent space: 📏 100) → synthetic images (🖼️ 3x64x64).

• 🔍 Discriminator (DNet) Receives real or generated images → outputs a probability of being "real" (✅ or ❌).

• 🔄 Training Loop

  • Discriminator: Learns to distinguish real 📸 vs. fake 🤖 images.
  • Generator: Learns to fool 🎭 the discriminator.
  • Optimizer: Adam (📈 lr=0.0002).

• 💾 Save & Reuse Models are saved as .pth files for later inference or fine-tuning.

🗺️ Schema

Here is the GAN architecture :

---

📂 Repository Structure

/file  
├── Dataset_image/   
│ ├── image_folder/chats            # Folder containing training images / Here, it's cat 😺 !  
│ ├── model/                        # Saved models (generator & discriminator)  
│ ├── result_image/                 # Generated results  
├── Img/                            # Subfolder with the example image for the README.md  
├── LICENSE  
├── README.md  
├── main.py

---

💻 Run it on Your PC

Clone the repository and install dependencies:

git clone https://github.com/Thibault-GAREL/Image_generator_GAN.git
cd Image_generator_GAN
pip install numpy torch torchvision matplotlib
# install with Cuda:
# pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

Train the GAN (adjust number of epochs 🔁 in the script):

python main.py

To generate and display images using the last models:

python main.py  #Put num_epochs = 0 l.40

---

📖 Inspiration / Sources

• 🎥 I follow the learning video : GAN from NeoCode

Code created by me 😎, Thibault GAREL - Github