Abstract
Radiance fields have gradually become a main representation of media.Although its appearance editing has been studied, how to achieveview-consistent recoloring in an efficient manner is still under explored. Wepresent RecolorNeRF, a novel user-friendly color editing approach for theneural radiance field. Our key idea is to decompose the scene into a set ofpure-colored layers, forming a palette. Thus, color manipulation can beconducted by altering the color components of the palette directly. To supportefficient palette-based editing, the color of each layer needs to be asrepresentative as possible. In the end, the problem is formulated as in anoptimization formula, where the layers and their blending way are jointlyoptimized with the NeRF itself. Extensive experiments show that ourjointly-optimized layer decomposition can be used against multiple backbonesand produce photo-realistic recolored novel-view renderings. We demonstratethat RecolorNeRF outperforms baseline methods both quantitatively andqualitatively for color editing even in complex real-world scenes.