Comic Colorization Project

A deep learning project to colorize grayscale comic images using a U-Net architecture.

---

Project Structure

├── API Key here/       # API key. You can borrow mine
├── Data/               # Data directory
│   ├── comic_dataset/  # Raw comic images
│   └── prepared_data/  # Preprocessed numpy arrays
├── Results/            # Generated result-images from the model 
├── Src/                # the model architecture and generate results
└── Notebooks/          # Jupyter notebooks (both loading data and training)

---

Setup & Installation

1. Clone the Repository

git clone https://github.com/Sebastianrix/CNN_U-Net_Project
cd CNN_U-Net_Project

2. Create and Activate Virtual Environment (optional but recommended)

python -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

---

GPU Setup

To enable GPU training with TensorFlow:

Requirements:

• CUDA 11.8
• cuDNN 8.6

Steps (Windows):

1. Install CUDA 11.8: https://developer.nvidia.com/cuda-11-8-0-download-archive
2. Download cuDNN 8.6: https://developer.nvidia.com/rdp/cudnn-download
3. Copy the contents of cuDNN's bin/, include/, and lib/ folders into your CUDA installation.
4. Add the CUDA bin/ and libnvvp/ paths to your Windows System Environment PATH variable.
5. Download zlibwapi.lib if missing, and place it in your CUDA lib/ directory.

---

Preprocess data & train the model

1. Preprocess the data

jupyter notebook Notebooks/LoadComicData_rgb.ipynb

2. Train the U-Net Model (RGB Version)

jupyter notebook Notebooks/train_unet_rgb_v2.ipynb

You can swap to LAB or HSV versions by running other notebooks, but are correctly broken / partcially unfucntional.

---