Hallucinated Neural Radiance Fields in the Wild

TL;DR

This paper introduces Ha-NeRF, an end-to-end framework for hallucinating realistic Neural Radiance Fields at different times of day from tourism images, effectively handling appearance changes and occlusions for view synthesis.

Contribution

The paper presents a novel appearance hallucination and anti-occlusion modules to improve view consistency and realism in hallucinated NeRFs, especially under complex occlusions and appearance variations.

Findings

Successfully hallucinated appearances in synthetic and real data.

Achieved occlusion-free view synthesis with improved realism.

Demonstrated effectiveness over existing methods.

Abstract

Neural Radiance Fields (NeRF) has recently gained popularity for its impressive novel view synthesis ability. This paper studies the problem of hallucinated NeRF: i.e., recovering a realistic NeRF at a different time of day from a group of tourism images. Existing solutions adopt NeRF with a controllable appearance embedding to render novel views under various conditions, but they cannot render view-consistent images with an unseen appearance. To solve this problem, we present an end-to-end framework for constructing a hallucinated NeRF, dubbed as Ha-NeRF. Specifically, we propose an appearance hallucination module to handle time-varying appearances and transfer them to novel views. Considering the complex occlusions of tourism images, we introduce an anti-occlusion module to decompose the static subjects for visibility accurately. Experimental results on synthetic data and real tourism photo collections demonstrate that our method can hallucinate the desired appearances and render occlusion-free images from different views. The project and supplementary materials are available at https://rover-xingyu.github.io/Ha-NeRF/.