PaletteNeRF: Palette-based Appearance Editing of Neural Radiance Fields

View #1 Style Transfer (c) Recoloring Illumination Editing (d) Applications View #2 View #1 View #2 Figure 1. We propose PaletteNeRF, a novel method for efficient appearance editing of neural radiance fields (NeRF). Taking (a) multi-view photos as training input, our approach reconstructs a NeRF and decomposes its appearance into a set of (b) 3D palette-based color bases. This enables (c) intuitive and photorealistic recoloring of the scene with 3D consistency across arbitrary views. Further, we show that (d) our method supports various palette-based editing applications such as illumination modification and 3D photorealistic style transfer.