Reproduce “OmniSSR: Zero-shot Omnidirectional Image Super-Resolution using Stable 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
Runyi Li, Xuhan Sheng, Weiqi Li, Jian Zhang
OmniSSR introduces the first zero-shot diffusion-based approach for omnidirectional image super-resolution, leveraging an off-the-shelf Stable Diffusion model. The method effectively adapts planar image priors to 360-degree images, demonstrating high fidelity and superior visual realness on benchmark datasets.
Key insights
- Powerful pre-trained planar generative models can be successfully adapted for complex, geometrically challenging tasks like ODISR in a zero-shot manner, eliminating the need for specialized training data.
- Iterative information exchange between global (ERP) and local (TP) representations during the generative process is crucial for mitigating projection distortions and ensuring global visual coherence in omnidirectional images.
- Integrating explicit data-fidelity constraints, such as Gradient Decomposition Correction, into diffusion models allows for an effective balance between generating realistic details and maintaining faithfulness to the degraded input.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.