NeuroStyle: Architectural Neural Style Transfer (NST)

NeuroStyle is a deep learning synthesis engine that decouples and recombines artistic style from photographic content. By utilizing a custom-built Convolutional Neural Network (CNN) with Residual Blocks, the system can "re-paint" any photograph using the textures and color profiles of 39+ distinct art movements.

🚀 Key Features

• Residual-Block Architecture: Employs a sophisticated CNN featuring Downsampling, Residual Skip-Connections for gradient stability, and Upsampling (Conv2DTranspose) for high-fidelity output generation.
• Dual-Feature Extraction:
  • Content Extraction: Captures semantic scenery and object arrangement from deep network layers ($x3, x6$).
  • Style Extraction: Captures texture correlations via Gram Matrices from multi-scale layers ($x1, x2, x5$).
• Vast Style Library: Includes specialized support for 39+ unique styles, including Street Art, Ukiyo-e, Suprematism, Bauhaus, and Surrealism.
• Latent Space Optimization: Utilizes the Adam Optimizer to iteratively evolve a noise-initialized tensor into a final masterpiece.

🛠️ Technologies Used

• TensorFlow 2.x: Core framework for model construction and tf.GradientTape optimization.
• Numpy & PIL: Used for image-to-tensor transformations and high-fidelity LANCZOS resampling.
• Matplotlib: For real-time training visualization and final rendering.

🧪 Methodology

Grounding its logic in the seminal research by Gatys et al., NeuroStyle operates on the principle that style and content are separable in the feature spaces of performance-optimized neural networks.

1. Gram Matrix Calculation: Computes the inner product between vectorized feature maps to capture texture while discarding global arrangement.
2. Loss Joint-Minimization: $$\mathcal{L}{total} = \alpha\mathcal{L}{content} + \beta\mathcal{L}_{style}$$
3. Backpropagation: The gradient of the total loss is used to iteratively update the pixel values of the generated image $\vec{x}$.

📂 Project Structure

NeuroStyle
├── ART_STYLES/      # 39+ distinct style images (Renaissance, Street Art, etc.)
├── reference_paper/ # Scientific foundations (Gatys_Image_Style_Transfer.pdf)
├── DL_PROJECT.ipynb # Main implementation and training pipeline
└── README.md        # Project documentation