N2F2: Hierarchical Scene Understanding with Nested Neural Feature Fields This paper introduces Nested Neural Feature Fields (N2F2), a novel approach for hierarchical scene understanding that employs hierarchical supervision to learn a single feature field, where different dimensions within the same high-dimensional feature encode scene properties at varying granularities. The method allows for a flexible definition of hierarchies, tailored to either the physical dimensions or semantics or both, enabling a comprehensive and nuanced understanding of scenes. The approach leverages a 2D class-agnostic segmentation model to provide semantically meaningful pixel groupings at arbitrary scales in the image space, and queries the CLIP vision-encoder to obtain language-aligned embeddings for each of these segments. The hierarchical supervision method then assigns different nested dimensions of the feature field to distill the CLIP embeddings using deferred volumetric rendering at varying physical scales, creating a coarse-to-fine representation. Extensive experiments show that the approach outperforms the state-of-the-art feature field distillation methods on tasks such as open-vocabulary 3D segmentation and localization, demonstrating the effectiveness of the learned nested feature field. The paper presents N2F2, which uses 3D Gaussian Splatting (3DGS) to represent the scene, augmented with a feature field that captures scene properties across different scales and semantic granularities. The method introduces a hierarchical supervision method that encodes varying levels of scene granularity into different subsets of the same feature space, leading to the idea of nested feature fields. A novel composite embedding approach is proposed that enables N2F2 to handle compound open-vocabulary queries during inference. The method is evaluated on challenging datasets, showing significant improvements in performance compared to existing methods. The approach is efficient and effective, achieving a 1.7× speedup over the current leading approach, LangSplat, with better accuracy and increased granularity. The paper also discusses the limitations of the method, including challenges in handling global scene context queries and the need for diverse image inputs to learn accurate feature fields. The method is shown to outperform existing approaches on complex compound and partitive constructions, such as "bag of cookies", "chair legs", and "blueberry donuts". The paper concludes that N2F2 significantly advances open-vocabulary 3D segmentation and localization tasks through its hierarchical supervision methodology.N2F2: Hierarchical Scene Understanding with Nested Neural Feature Fields This paper introduces Nested Neural Feature Fields (N2F2), a novel approach for hierarchical scene understanding that employs hierarchical supervision to learn a single feature field, where different dimensions within the same high-dimensional feature encode scene properties at varying granularities. The method allows for a flexible definition of hierarchies, tailored to either the physical dimensions or semantics or both, enabling a comprehensive and nuanced understanding of scenes. The approach leverages a 2D class-agnostic segmentation model to provide semantically meaningful pixel groupings at arbitrary scales in the image space, and queries the CLIP vision-encoder to obtain language-aligned embeddings for each of these segments. The hierarchical supervision method then assigns different nested dimensions of the feature field to distill the CLIP embeddings using deferred volumetric rendering at varying physical scales, creating a coarse-to-fine representation. Extensive experiments show that the approach outperforms the state-of-the-art feature field distillation methods on tasks such as open-vocabulary 3D segmentation and localization, demonstrating the effectiveness of the learned nested feature field. The paper presents N2F2, which uses 3D Gaussian Splatting (3DGS) to represent the scene, augmented with a feature field that captures scene properties across different scales and semantic granularities. The method introduces a hierarchical supervision method that encodes varying levels of scene granularity into different subsets of the same feature space, leading to the idea of nested feature fields. A novel composite embedding approach is proposed that enables N2F2 to handle compound open-vocabulary queries during inference. The method is evaluated on challenging datasets, showing significant improvements in performance compared to existing methods. The approach is efficient and effective, achieving a 1.7× speedup over the current leading approach, LangSplat, with better accuracy and increased granularity. The paper also discusses the limitations of the method, including challenges in handling global scene context queries and the need for diverse image inputs to learn accurate feature fields. The method is shown to outperform existing approaches on complex compound and partitive constructions, such as "bag of cookies", "chair legs", and "blueberry donuts". The paper concludes that N2F2 significantly advances open-vocabulary 3D segmentation and localization tasks through its hierarchical supervision methodology.