Jun Xu

Neural Radiance Fields (NeRFs) have gained popularity by demonstrating impressive capabilities in synthesizing realistic novel views from multiple view images. It approximates the continuous integration of rays as a finite Riemann sum of the estimated colors and densities of the sampled points. Although this allows for efficient rendering, approximating divergent integrals under varying directions and interval lengths with piecewise constant features does not account for high-order variations within integration intervals, leading to ambiguous representations and limited reconstruction quality. In this paper, we propose to model the distribution within each interval using explicit Taylorlike explicit polynomials parameterized by spatial gradients. These polynomials encode the length and direction information of integrals, helping to disambiguate interval distributions and mitigate integral approximation errors in volume rendering. Our framework includes a learnable gradient estimator, regularized via an interval boundary loss to ensure numerical consistency and smoothness, alongside an adaptive interval scaling module that enhances stability and handles varying interval sizes. Our proposed method allows an easy integration with existing NeRFbased rendering frameworks. Experimental results on both synthetic and real-world scenes demonstrate that our method significantly boosts the rendering quality of various NeRF models, achieving state-of-the-art performance.

Neural radiance fields (NeRFs) have achieved impressive view synthesis results by learning an implicit volumetric representation from multi-view images. To project the implicit representation into an image, NeRF employs volume rendering that approximates the continuous integrals of rays as an accumulation of the colors and densities of the sampled points. Although this approximation enables efficient rendering, it ignores the direction information in point intervals, resulting in ambiguous features and limited reconstruction quality. In this paper, we propose a learning method that utilizes learnable view-dependent features to improve scene representation and reconstruction. We model the volume rendering integral with a piecewise constant volume density and spherical harmonic-guided view-dependent features, facilitating ambiguity elimination while preserving the rendering efficiency. In addition, we introduce a regularization term that restricts the anisotropic representation effect to be local, with negligible effect on geometry representations, and that encourages recovering the correct geometry. Our method is flexible and can be plugged into NeRF-based frameworks. Extensive experiments show that the proposed representation can boost the rendering quality of various NeRFs and achieve state-of-the-art rendering performance on both synthetic and real-world scenes.