A collection of recent methods on 3D generation from text description. There are mainly 2 kinds of methods of text-to-3D generation:
- Direct End-to-End Generation
(There are multiple internal steps, but they are transparent to the user)
- initialize a coarse layout from text, and then refine/inpaint it
- generate a local scene from text, and then outpaint/optimize it
- Sequential Multi-Stage Generation
(Each internal step has an independent output as the input for the next stage)
- reconstruction based on text-to-image models and depth-estimation models
- reconstruct based on the multi-view generation models from text
- reconstruct a premitive scene from a text-to-image model, then gradually expand it and align features
This repo focuses on the Sequential Multi-Stage Generation approach, and the generation about 3D scene. As for the other topic, please refer to the comprehensive collections listed under Related-Repos-and-Websites at the end of this file. Feel free to submit a pull request if you have relevant papers to add.
Other repos:
- Text-to-3D for a carefully compiled collection of text-to-3D research papers.
- Awesome Text-to-3D for a curated list of text-to-3D.
About abbreviation: In the list below: B for best paper, S for spotlight, H for highlight, W for workshop.
- 2021.06 - Text2Mesh: Text-Driven Neural Stylization for Meshes (CVPR 2022) : This work develops intuitive controls for editing the style of 3D objects by predicting color and local geometric details based on a target text prompt.
- 2021.12 - DreamField: Zero-Shot Text-Guided Object Generation with Dream Fields (CVPR 2022) : This paper utilizes a two-stage optimization framework to create high-quality 3D mesh models in a shorter time.
- 2022.09 - DreamFusion: Text-to-3D using 2D Diffusion (ICLR 2023) : This paper introduces a method for generating 3D objects using 2D diffusion models.
- 2022.11 - Magic3D: High-Resolution Text-to-3D Content Creation (CVPR 2023) : This paper introduces a method for generating 3D objects using 2D diffusion models.
- 2023.03 - Fantasia3D: Disentangling Geometry and Appearance for High-quality Text-to-3D Content Creation (ICCV 2023) : This research focuses on disentangling geometry and appearance for high-quality 3D content creation.
- 2023.05 - HiFA: High-fidelity Text-to-3D Generation with Advanced Diffusion Guidance (ICLR 2024) : This paper proposes holistic sampling and smoothing approaches to achieve high-quality text-to-3D generation in a single-stage optimization.
- 2023.10 - GaussianDreamer: Fast Generation from Text to 3D Gaussians by Bridging 2D and 3D Diffusion Models (CVPR 2024) : This paper introduces a novel framework designed to efficiently produce high-quality 3D assets from textual prompts.
- 2023.11 - LucidDreamer: Towards High-Fidelity Text-to-3D Generation via Interval Score Matching (CVPR 2024) (H) : This research introduces a novel method called Interval Score Matching (ISM) for generating high-fidelity 3D models.
- 2023.11 - LucidDreamer: Domain-free Generation of 3D Gaussian Splatting Scenes (arXiv 2023) (1.3k stars!) : This research introduces a novel method called Interval Score Matching (ISM) for generating high-fidelity 3D models.
- 2024.02 - GALA3D: Towards Text-to-3D Complex Scene Generation via Layout-guided Generative Gaussian Splatting (ICML 2024) : This research introduces a novel framework for generating complex 3D scenes from textual descriptions.
- 2024.04 - RealmDreamer: Text-Driven 3D Scene Generation with Inpainting and Depth Diffusion (arXiv 2024) : This research introduces a model to use pretrained inpainting and depth priors with a robust initialization of a 3D Gaussian Splatting model.
- 2024.06 - GradeADreamer: Enhanced Text-to-3D Generation Using Gaussian Splatting and Multi-View Diffusion (arXiv 2024) : This research introduces a novel three-stage training pipeline called GradeADreamer, which aims to address common challenges in text-to-3D generation, such as the Multi-face Janus problem and extended generation time for high-quality assets.
- 2024.07 - PlacidDreamer: Advancing Harmony in Text-to-3D Generation (ACM MM 2024) : This research explores methods for multi-view consistency and detail optimization.
- 2024.07 - ScaleDreamer: Scalable Text-to-3D Synthesis with Asynchronous Score Distillation (ECCV 2024) : This paper introduces an asynchronous score distillation method to enhance generation quality.
- 2024.08 - DreamLCM: Towards High-Quality Text-to-3D Generation via Latent Consistency Model (ACM MM 2024) : This paper proposes a method to improve 3D generation quality through a latent consistency model.
- 2020.02 - CDISN: Deep Implicit Surface Network for High-quality Single-view 3D Reconstruction (CVPR 2020)
- 2020.03 - 3D Photography using Context-aware Layered Depth Inpainting (CVPR 2020)
- 2020.03 - Learning Implicit Fields for Generative Shape Modeling (CVPR 2020)
- 2020.03 - Pix2Vox++: Multi-Scale Context-Aware 3D Object Reconstruction from Single and Multiple Images (CVPR 2020)
- 2020.03 - Neural Mesh Flow: 3D Manifold Mesh Generation via Diffeomorphic Flows (CVPR 2020)
- 2020.03 - Neural Scene Flow Fields for Space-Time View Synthesis of Dynamic Scenes (CVPR 2020)
- 2021.03 - Text2Shape: Generating Shapes from Natural Language by Learning Joint Embedding (CVPR 2021)
- 2021.06 - Text2Mesh: Text-Driven Neural Stylization for Meshes (SIGGRAPH 2021)
- 2021.09 - CLIP-Forge: Towards Zero-Shot Text-to-Shape Generation (NeurIPS 2021)
- 2022.09 - DreamFusion: Text-to-3D using 2D Diffusion (ICLR 2023)
- 2022.11 - Magic3D: High-Resolution Text-to-3D Content Creation (CVPR 2023)
- 2023.03 - Fantasia3D: Disentangling Geometry and Appearance for High-quality Text-to-3D Content Creation (ICCV 2023)
- 2023.05 - HiFA: High-fidelity Text-to-3D Generation with Advanced Diffusion Guidance (ICLR 2024)
- 2023.08 - IT3D: Improved Text-to-3D Generation with Explicit View Synthesis (AAAI 2024)
- 2023.10 - GaussianDreamer: Fast Generation from Text to 3D Gaussians by Bridging 2D and 3D Diffusion Models (CVPR 2024)
- 2023.11 - LucidDreamer: Towards High-Fidelity Text-to-3D Generation via Interval Score Matching (CVPR 2024) (H) : This research introduces a novel method called Interval Score Matching (ISM) for generating high-fidelity 3D models.
- 2023.11 - LucidDreamer: Domain-free Generation of 3D Gaussian Splatting Scenes (arXiv 2023) (1.3k stars!) : This research introduces a novel method called Interval Score Matching (ISM) for generating high-fidelity 3D models.
- 2023.12 - Sherpa3D: Boosting High-Fidelity Text-to-3D Generation via Coarse 3D Prior (CVPR 2024)
- 2024.02 - GALA3D: Towards Text-to-3D Complex Scene Generation via Layout-guided Generative Gaussian Splatting (ICML 2024)
- 2024.04 - RealmDreamer: Text-Driven 3D Scene Generation with Inpainting and Depth Diffusion (arXiv 2024)
- 2024.06 - GradeADreamer: Enhanced Text-to-3D Generation Using Gaussian Splatting and Multi-View Diffusion （arXiv 2024)
- 2024.07 - PlacidDreamer: Advancing Harmony in Text-to-3D Generation (ACM MM 2024)
- 2024.07 - ScaleDreamer: Scalable Text-to-3D Synthesis with Asynchronous Score Distillation (ECCV 2024)
- 2024.08 - DreamLCM: Towards High-Quality Text-to-3D Generation via Latent Consistency Model (ACM MM 2024)
- 2024.08 - SceneDreamer360: Text-Driven 3D-Consistent Scene Generation with Panoramic Gaussian Splatting (arXiv 2024)
- 2024.08 - LayerPano3D: Layered 3D Panorama for Hyper-Immersive Scene Generation (arXiv 2024)
- 2024.02 WonderJourney: Going from Anywhere to Everywhere (CVPR 2024)
- 2024.04 PhysDreamer: Physics-Based Interaction with 3D Objects via Video Generation (ECCV 2024)