Multi-style Neural Radiance Field with AdaIN

TL;DR

The paper tackles stylized novel view synthesis for 3D scenes by learning a single model capable of rendering multiple artistic styles. It proposes a three-stage pipeline that injects AdaIN style statistics into the NeRF representation, enabling voxel-wise style transfer during rendering; AdaIN aligns content/style statistics via channel-wise means and variances, i.e., $ \mu$ and $ \sigma^2$, and these statistics are incorporated into the final NeRF MLPs. Key contributions include a simplified multi-style pipeline, a density-aware extension to better handle styles with strong brush strokes, and a style interpolation mechanism to balance content and style. Together these enable flexible, controllable stylized rendering for 3D scenes with practical implications for VR/AR content creation.

Abstract

In this work, we propose a novel pipeline that combines AdaIN and NeRF for the task of stylized Novel View Synthesis. Compared to previous works, we make the following contributions: 1) We simplify the pipeline. 2) We extend the capabilities of model to handle the multi-style task. 3) We modify the model architecture to perform well on styles with strong brush strokes. 4) We implement style interpolation on the multi-style model, allowing us to control the style between any two styles and the style intensity between the stylized output and the original scene, providing better control over the stylization strength.

Figures (2)

• Figure 1: Multi-style NeRF
• Figure 2: Multi-style NeRF with Density Aware