Scholarlink

Neural Radiance Fields (NeRFs) have significantly advanced Novel View Synthesis (NVS) by implicitly representing 3D scenes through the weights of a neural network. However, the practical applicability of NeRFs are critically dependent on the quality and consistency of input data, typically requiring clean images with known camera poses, controlled lighting, and static content. In this work, we simulate and analyze the effects of different types of image degradations on NeRF reconstruction quality across various scene configurations. Furthermore, we propose and evaluate Pre-NeRF, a preprocessing pipeline incorporating a suite of algorithms designed to mitigate the impact of these degradations on input imagery. Our experimental results demonstrate quantifiable improvements in NeRF reconstruction fidelity when utilizing the proposed preprocessing pipeline, suggesting its potential value for enhancing NeRF performance in less-than-ideal real-world conditions.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Pre-NeRF: Evaluating Preprocessing Approaches to Mitigate Real-World Corruptions in NeRF Reconstruction

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Alexander Suen,
  • Cayden Gu

摘要

Neural Radiance Fields (NeRFs) have significantly advanced Novel View Synthesis (NVS) by implicitly representing 3D scenes through the weights of a neural network. However, the practical applicability of NeRFs are critically dependent on the quality and consistency of input data, typically requiring clean images with known camera poses, controlled lighting, and static content. In this work, we simulate and analyze the effects of different types of image degradations on NeRF reconstruction quality across various scene configurations. Furthermore, we propose and evaluate Pre-NeRF, a preprocessing pipeline incorporating a suite of algorithms designed to mitigate the impact of these degradations on input imagery. Our experimental results demonstrate quantifiable improvements in NeRF reconstruction fidelity when utilizing the proposed preprocessing pipeline, suggesting its potential value for enhancing NeRF performance in less-than-ideal real-world conditions.