msracver/Deep-Image-Analogy

Establishes dense semantic correspondences between image pairs using VGG-19 deep features combined with PatchMatch, enabling style transfer across diverse image types (photo-to-painting, sketch-to-photo, etc.). Built as a C++ CUDA implementation on Caffe, it extracts multi-scale CNN features and performs weighted blending with optional edge-preserving filtering for structure-aware synthesis. Supports GPU acceleration on NVIDIA hardware with configurable parameters for blend strength and resolution scaling to balance fidelity and computational cost.

1,374 stars. No commits in the last 6 months.