Reproduce “Optimal Image Transport on Sparse Dictionaries” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Junqing Huang, Haihui Wang, Andreas Weiermann, Michael Ruzhansky
The paper presents an "Optimal Image Transport on Sparse Dictionaries" framework, integrating sparse representation and optimal transport for image-to-image translation, yielding high-quality color transformations and photo-realistic style transfers. It demonstrates superior structural fidelity (SSIM 0.7934) and perceptual similarity (LPIPS 0.2146) while mitigating artifacts common in traditional optimal transport and deep learning approaches.
Key insights
- Applying optimal transport to compact, learned sparse dictionaries significantly reduces the computational burden compared to applying it directly to raw image data, making it empirically solvable.
- A unified framework for simultaneously learning image representations and performing feature transformations leads to more consistent and higher-quality image-to-image translation outcomes.
- The method provides an interpretable and robust alternative to deep learning, demonstrating superior preservation of structural fidelity and local textures, especially in photo-realistic applications.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.