🔮 Image-Color-Enhancement with U-Net and Attention U-Net

This project implements deep learning models for automatic color correction of RAW images, aiming to enhance image quality by mapping unprocessed RAW inputs to professionally corrected outputs.

🚀 Features

Custom U-Net and Attention U-Net architectures for image-to-image translation.

Loss Functions:

• L1 Loss (pixel-level reconstruction).

• Perceptual Loss using pretrained VGG16 (captures semantic similarity).

Training Enhancements:

• Dropout regularization.

• Early Stopping with patience.

Visualization:

• Compare RAW Input, Model Prediction, and Ground Truth during training.

📂 Dataset

This project uses the Adobe FiveK Dataset Dataset Link, which provides RAW images and corresponding corrected images by professional photographers.

• 5000 RAW images → Input.

• 5000 Corrected images → Ground Truth target.

🏗 Model Architectures

🔹 U-Net

• Encoder-Decoder structure with skip connections.

• Captures both low-level and high-level features for effective reconstruction.

🔹 Attention U-Net

• Enhances U-Net with attention gates in skip connections.

• Model focuses on relevant regions of the image → cleaner and more accurate predictions.

⚙️ Installation

• Clone the repo and install dependencies:

  git clone https://github.com/HabibaMAtiia/Image-Color-Enhancement.git
  cd Image-Color-Enhancement
  pip install -r requirements.txt
  

📥 Pretrained Model

• You can download the trained model from Google Drive: Download Model Weights

📊 Example Result

✅ Future Improvements

• Experiment with GAN-based approaches (e.g., Pix2Pix, CycleGAN).
• Build a simple web app using frameworks like Gradio.
• Add evaluation metrics: PSNR, SSIM.

👤 Author

Developed by Habiba Mohammad: 📩 habibamohamad062@gmail.com