jcjohnson/neural-style
Neural Style
This is a PyTorch implementation of a neural style transfer system. It functions as a convolutional neural network image stylizer and artistic style blender designed to combine the content of one image with the artistic style of another.
The system supports blending multiple style sources and adjusting the relative weights between content and style reconstruction. It includes capabilities for preserving the original color palette of the content image and adjusting style scales to determine which artistic patterns are transferred.
The pipeline enables high-resolution image processing by distributing neural network layers across multiple graphics cards.
Features
- CNN Image Stylizers - Utilizes convolutional neural networks to extract artistic patterns and apply them to photographs.
- Feature Extraction - Uses a pre-trained VGG-19 network to extract hierarchical visual features and spatial structures.
- Neural Style Transfer - Provides a system for neural style transfer that allows custom control over style scale and color preservation.
- Content and Style Balancing - A feature in the tool that adjusts the relative weights between content and style reconstruction to control visual influence in the final image.
- Gradient Descent Algorithms - Employs iterative gradient descent to refine the output image by minimizing the content and style loss functions.
- Gram Matrix Style Representations - Implements Gram matrix calculations to extract and represent artistic textures from style images.
- Multi-GPU Distribution - Splits heavy neural network computations across multiple graphics cards for high-resolution image synthesis.
- Multi-Style Blending - Enables combining artistic elements from several style images by assigning relative weights to each.
- Style Scale Adjustment - Resizes the style image before extraction to control which specific artistic patterns are applied to the content.
- Static Loss Weighting - Provides adjustable coefficients to balance the influence of content and style reconstruction.
- Multi-GPU Layer Distribution - Distributes neural network layers across multiple GPUs to enable the processing of high-resolution images.
- Color Preservation - A feature in the tool that maintains the color palette of the original content image while applying the artistic patterns of a style image.
- High-Performance Image Pipelines - Implements a high-performance image pipeline that distributes processing across multiple GPUs for high resolution.
- Color Space Converters - Performs color space conversions to preserve the original hues of the content image during stylization.
- Artistic Style Transfer - Neural algorithm for transferring artistic styles between images.
- Machine Learning and AI - Implementation of neural style transfer algorithms.
