Fast Omni-Directional Image Super-Resolution: Adapting the Implicit Image Function with Pixel and Semantic-Wise Spherical Geometric Priors

TL;DR

ODIs suffer from ERP-induced distortion, complicating SR while demanding fast, flexible upscaling. The paper presents FAOR, a fast arbitrary-scale ODI-SR framework that adapts the implicit image function to ERP by injecting spherical priors into both the latent representation (SAFE with ATFM and CA) and the reconstruction process (SGIF with geodesic resampling). FAOR achieves superior performance and faster inference than state-of-the-art ODI-SR methods, as demonstrated on ODI-SR and SUN360 with ablations confirming the contributions of the priors and the geodesic resampling strategy. This approach offers a practical pathway to real-time, high-quality ODI upscaling without heavy spherical convolutions or reprojection overheads.

Abstract

In the context of Omni-Directional Image (ODI) Super-Resolution (SR), the unique challenge arises from the non-uniform oversampling characteristics caused by EquiRectangular Projection (ERP). Considerable efforts in designing complex spherical convolutions or polyhedron reprojection offer significant performance improvements but at the expense of cumbersome processing procedures and slower inference speeds. Under these circumstances, this paper proposes a new ODI-SR model characterized by its capacity to perform Fast and Arbitrary-scale ODI-SR processes, denoted as FAOR. The key innovation lies in adapting the implicit image function from the planar image domain to the ERP image domain by incorporating spherical geometric priors at both the latent representation and image reconstruction stages, in a low-overhead manner. Specifically, at the latent representation stage, we adopt a pair of pixel-wise and semantic-wise sphere-to-planar distortion maps to perform affine transformations on the latent representation, thereby incorporating it with spherical properties. Moreover, during the image reconstruction stage, we introduce a geodesic-based resampling strategy, aligning the implicit image function with spherical geometrics without introducing additional parameters. As a result, the proposed FAOR outperforms the state-of-the-art ODI-SR models with a much faster inference speed. Extensive experimental results and ablation studies have demonstrated the effectiveness of our design.

Figures (4)

• Figure 1: (a) The proposed FAOR provides a continuous ODI representation, enabling arbitrary-scale SR. (b) Comparisons with other state-of-the-art models regarding performance and inference speed on the $\times 8$ SR task
• Figure 2: Overall architectures of the proposed FAOR and detailed design of key modules.
• Figure 3: (a) The normalized interpolation method in LIIF, leveraging the areas of $s_0$, $s_1$, $s_2$, and $s_3$ to obtain normalized weights for $z_p$ interpolation. (b) The proposed spherical geodesic-based interpolation.
• Figure 4: Visual comparisons of $\times 8$ SR results on ODI-SR and SUN360 testing sets.