This one of my assignments for Deep Learning. In this project, we are tasked with implementing our own version based on the Neural Style Transfer tutorial from TensorFlow. tutorial
My primary objective is to generate mosaic windows featuring unique motifs that influenced by modern elements,
The Future is then to convert these generated images into SVG files. These SVGs can later be used in applications like Fusion 360 (or similar software) to extrude the designs, thereby creating files that can be 3D printed, laser cut, or manufactured by other means.
Note: A Fusion 360 student account is provided free of charge.
Neural Style Transfer
This upcoming version is still a work in progress and currently meets the basic requirement of the assignment by focusing on the implementation aspect. git project.
Colors are not really necessary for my model. However, if you aim to create this mosaic window, a colorful vision might be appreciated. The next step will involve combining black-and-white mosaic images with colored content images to focus on the essentials. For later, with 1(B&W) instead of 3 dimensions(RGB)
How to Use
The Jupyter Notebook can be downloaded and, once all required libraries are installed, executed without issues.
To update the images, simply specify a different path in the configuration. The comments serve as both documentation and explanation of the code. You can also modify other configuration settings as needed.
Please note that if you plan to push this further, it is recommended to use a GPU. If you're using Nvidia hardware, refer to this link for more details: Nvidia CUDA Toolkit.
Examples
Conclusion for my Goal
To create mosaic windows, I first tried a direct approach with my Configurator. However, I believe it would be more effective to apply filters or algorithms that adjust the edges to form intuitive shapes. This method will separately detect outlines and use them as a grid to build the mosaic.