Reproduce “Learning Deblurring Texture Prior from Unpaired Data with Diffusion Model” 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
Chengxu Liu, Lu Qi, Jinshan Pan, Xueming Qian, Ming-Hsuan Yang
TP-Diff introduces a diffusion model-based framework for unsupervised image deblurring, learning spatially varying texture priors from unpaired blurry and sharp images. This method achieves state-of-the-art performance on various benchmarks while maintaining computational efficiency and improved generalization to real-world blur.
Key insights
- Diffusion Models can effectively learn and generate robust, spatially varying texture priors for unpaired image restoration, demonstrating a novel application domain for DMs.
- Disentangling the learning of texture priors from unpaired data (using a specialized encoder) and their generative modeling (using a Diffusion Model) leads to a robust and effective unsupervised deblurring solution.
- The memory-enhanced Texture Prior Encoder (TPE) can effectively extract relevant texture information from unpaired sharp images to guide deblurring of blurry images.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.