Feature 3DGS: Supercharging 3D Gaussian Splatting to Enable Distilled Feature Fields

• URL: http://arxiv.org/abs/2312.03203v3
• Date: Mon, 8 Apr 2024 07:19:52 GMT
• Title: Feature 3DGS: Supercharging 3D Gaussian Splatting to Enable Distilled Feature Fields
• Authors: Shijie Zhou, Haoran Chang, Sicheng Jiang, Zhiwen Fan, Zehao Zhu, Dejia Xu, Pradyumna Chari, Suya You, Zhangyang Wang, Achuta Kadambi,
• Abstract summary: Methods that use Neural Radiance fields are versatile for traditional tasks such as novel view synthesis. 3D Gaussian splatting has shown state-of-the-art performance on real-time radiance field rendering. We propose architectural and training changes to efficiently avert this problem.
• Score: 54.482261428543985
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 3D scene representations have gained immense popularity in recent years. Methods that use Neural Radiance fields are versatile for traditional tasks such as novel view synthesis. In recent times, some work has emerged that aims to extend the functionality of NeRF beyond view synthesis, for semantically aware tasks such as editing and segmentation using 3D feature field distillation from 2D foundation models. However, these methods have two major limitations: (a) they are limited by the rendering speed of NeRF pipelines, and (b) implicitly represented feature fields suffer from continuity artifacts reducing feature quality. Recently, 3D Gaussian Splatting has shown state-of-the-art performance on real-time radiance field rendering. In this work, we go one step further: in addition to radiance field rendering, we enable 3D Gaussian splatting on arbitrary-dimension semantic features via 2D foundation model distillation. This translation is not straightforward: naively incorporating feature fields in the 3DGS framework encounters significant challenges, notably the disparities in spatial resolution and channel consistency between RGB images and feature maps. We propose architectural and training changes to efficiently avert this problem. Our proposed method is general, and our experiments showcase novel view semantic segmentation, language-guided editing and segment anything through learning feature fields from state-of-the-art 2D foundation models such as SAM and CLIP-LSeg. Across experiments, our distillation method is able to provide comparable or better results, while being significantly faster to both train and render. Additionally, to the best of our knowledge, we are the first method to enable point and bounding-box prompting for radiance field manipulation, by leveraging the SAM model. Project website at: https://feature-3dgs.github.io/

Related papers

• 3D Convex Splatting: Radiance Field Rendering with 3D Smooth Convexes [87.01284850604495]
  We introduce 3D Convexting (3DCS), which leverages 3D smooth convexes as primitives for modeling geometrically-meaningful radiance fields from multiview images. 3DCS achieves superior performance over 3DGS on benchmarks such as MipNeizer, Tanks and Temples, and Deep Blending. Our results highlight the potential of 3D Convexting to become the new standard for high-quality scene reconstruction.
  arXiv  Detail & Related papers  (2024-11-22T14:31:39Z)
• DistillNeRF: Perceiving 3D Scenes from Single-Glance Images by Distilling Neural Fields and Foundation Model Features [65.8738034806085]
  DistillNeRF is a self-supervised learning framework for understanding 3D environments in autonomous driving scenes. Our method is a generalizable feedforward model that predicts a rich neural scene representation from sparse, single-frame multi-view camera inputs.
  arXiv  Detail & Related papers  (2024-06-17T21:15:13Z)
• 3D-HGS: 3D Half-Gaussian Splatting [5.766096863155448]
  Photo-realistic 3D Reconstruction is a fundamental problem in 3D computer vision. We propose to employ 3D Half-Gaussian (3D-HGS) kernels, which can be used as a plug-and-play kernel.
  arXiv  Detail & Related papers  (2024-06-04T19:04:29Z)
• ${M^2D}$NeRF: Multi-Modal Decomposition NeRF with 3D Feature Fields [33.168225243348786]
  We present a single model, Multi-Modal Decomposition NeRF ($M2D$NeRF), that is capable of both text-based and visual patch-based edits. Specifically, we use multi-modal feature distillation to integrate teacher features from pretrained visual and language models into 3D semantic feature volumes. Experiments on various real-world scenes show superior performance in 3D scene decomposition tasks compared to prior NeRF-based methods.
  arXiv  Detail & Related papers  (2024-05-08T12:25:21Z)
• Learning Naturally Aggregated Appearance for Efficient 3D Editing [94.47518916521065]
  We propose to replace the color field with an explicit 2D appearance aggregation, also called canonical image. To avoid the distortion effect and facilitate convenient editing, we complement the canonical image with a projection field that maps 3D points onto 2D pixels for texture lookup. Our representation, dubbed AGAP, well supports various ways of 3D editing (e.g., stylization, interactive drawing, and content extraction) with no need of re-optimization.
  arXiv  Detail & Related papers  (2023-12-11T18:59:31Z)
• PonderV2: Pave the Way for 3D Foundation Model with A Universal Pre-training Paradigm [114.47216525866435]
  We introduce a novel universal 3D pre-training framework designed to facilitate the acquisition of efficient 3D representation. For the first time, PonderV2 achieves state-of-the-art performance on 11 indoor and outdoor benchmarks, implying its effectiveness.
  arXiv  Detail & Related papers  (2023-10-12T17:59:57Z)
• FeatureNeRF: Learning Generalizable NeRFs by Distilling Foundation Models [21.523836478458524]
  Recent works on generalizable NeRFs have shown promising results on novel view synthesis from single or few images. We propose a novel framework named FeatureNeRF to learn generalizable NeRFs by distilling pre-trained vision models. Our experiments demonstrate the effectiveness of FeatureNeRF as a generalizable 3D semantic feature extractor.
  arXiv  Detail & Related papers  (2023-03-22T17:57:01Z)
• 3D-aware Image Synthesis via Learning Structural and Textural Representations [39.681030539374994]
  We propose VolumeGAN, for high-fidelity 3D-aware image synthesis, through explicitly learning a structural representation and a textural representation. Our approach achieves sufficiently higher image quality and better 3D control than the previous methods.
  arXiv  Detail & Related papers  (2021-12-20T18:59:40Z)
• Generative Occupancy Fields for 3D Surface-Aware Image Synthesis [123.11969582055382]
  Generative Occupancy Fields (GOF) is a novel model based on generative radiance fields. GOF can synthesize high-quality images with 3D consistency and simultaneously learn compact and smooth object surfaces.
  arXiv  Detail & Related papers  (2021-11-01T14:20:43Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.