Interactive Color Manipulation in NeRF: A Point Cloud and Palette-Driven Approach

Neural Radiance Fields (NeRF) technology has gained popularity for its ability to synthesize photorealistic new views of complex scenes. However, its editability is severely restricted due to its implicit representation. Existing color editing approaches often result in color contamination and distortion issues when editing NeRF-represented appearances. Additionally, these methods typically only support scene-level editing and cannot perform more fine-grained editing on objects within the scene. In this paper, we propose a fine-grained controllable color editing method that supports object-level operations while reducing the color contamination and distortion issues. Our method decomposes the color of each point in the scene into a linear combination of a set of palette bases. To ensure the sparsity of the decomposition, we propose Enhanced Sparse Regularizer (ESR) during the optimization process. We also propose a color correction function that reduces the error between the rendered color and the real color. Furthermore, we extend our model to support finer-grained local color editing through point cloud-level processing. Extensive experiments demonstrate that our color editing method outperforms baseline methods in terms of both qualitative and quantitative results.