EdgeNeRF: Edge-Guided Regularization for Neural Radiance Fields from Sparse Views

TL;DR

EdgeNeRF tackles the artifact-prone nature of sparse-view NeRF by introducing edge-guided, local regularization that preserves geometric boundaries. By extracting edges with DexiNed and applying depth and normal smoothing only in non-edge regions, EdgeNeRF achieves sharper boundaries and more consistent geometry, demonstrated by PSNR gains on LLFF and DTU over RegNeRF. The method is lightweight, patch-based, and plug-and-play, enabling easy integration with other sparse-view frameworks to boost performance without substantial training time. Limitations include difficulties with complex textures and additional computational cost from normal regularization, suggesting future work on semantic-aware regularization.

Abstract

Neural Radiance Fields (NeRF) achieve remarkable performance in dense multi-view scenarios, but their reconstruction quality degrades significantly under sparse inputs due to geometric artifacts. Existing methods utilize global depth regularization to mitigate artifacts, leading to the loss of geometric boundary details. To address this problem, we propose EdgeNeRF, an edge-guided sparse-view 3D reconstruction algorithm. Our method leverages the prior that abrupt changes in depth and normals generate edges. Specifically, we first extract edges from input images, then apply depth and normal regularization constraints to non-edge regions, enhancing geometric consistency while preserving high-frequency details at boundaries. Experiments on LLFF and DTU datasets demonstrate EdgeNeRF's superior performance, particularly in retaining sharp geometric boundaries and suppressing artifacts. Additionally, the proposed edge-guided depth regularization module can be seamlessly integrated into other methods in a plug-and-play manner, significantly improving their performance without substantially increasing training time. Code is available at https://github.com/skyhigh404/edgenerf.

Figures (7)

• Figure 1: Example of novel view synthesis from sparse views. Our edge-guided regularization yields sharper boundaries and more consistent geometry than state-of-the-art RegNeRFniemeyer2021regnerfregularizingneuralradiance.
• Figure 2: The pipeline of edge-guided sparse-view 3D reconstruction framework EdgeNeRF.
• Figure 3: Comparison on different edge extraction methods. (b) and (c) display binarized outputs from edge extraction. While the Canny operator fails to capture low-contrast edges (e.g., the back of the chair), DexiNed provides more complete and coherent edge maps, especially for fine structures.
• Figure 4: Global depth smoothing of RegNeRF causes edge blurring and reconstruction failure.
• Figure 5: Visual comparisons on the LLFF dataset with three input views. EdgeNeRF demonstrates superior performance in handling complex lighting conditions (flower scene) and fine geometric details (T-Rex model), with +0.41dB PSNR improvement over RegNeRF.