A curated list of awesome AIGC 3D papers
Toward Production-Ready 3D Asset Generation
Jiafeng Wu1,2*† · Zhuofan Lou1,3*† · Jian Liu1‡
Chunchao Guo4 · Dazhao Du1 · Song Guo1§
1 The Hong Kong University of Science and Technology
2 Huazhong University of Science and Technology · 3 Sichuan University · 4 Tencent
✦ Production-Oriented 3D Generation Survey ✦
Taxonomy · Data · Objects · Characters · Scenes · Evaluation · Industry
Three-dimensional content generation has progressed from producing isolated, visually plausible shapes to constructing structured assets that can be deployed in real-time interactive environments. This trajectory is driven by converging demands from game development, embodied AI, world simulation, digital twins, and spatial computing, all of which require 3D content that goes beyond surface appearance to satisfy engine-level constraints on topology, UV parameterization, physically based materials, skeletal rigging, and physics-aware scene layout. Despite rapid advances in generative modeling, a persistent gap separates the outputs of current methods from the production-ready standard expected by interactive applications. This survey addresses that gap by organizing the literature around the asset production pipeline rather than algorithmic families.
At a Glance
- Organized by a production-ready, pipeline-first taxonomy rather than isolated algorithm families.
- Covers three asset tiers: general objects, characters and avatars, and scenes and environments.
- Tracks the full asset workflow from data foundations through geometry, topology, appearance, rigging, and scene assembly.
- Consolidates methods, datasets, evaluation criteria, and industry references in one companion list.
The survey is organized around a two-dimensional taxonomy:
This structure mirrors the production pipeline used in game engines and interactive applications, enabling direct assessment of where each method fits within a deployment workflow.
| Dataset | Year | Scale | Description |
|---|---|---|---|
| ShapeNet | 2015 | 51K models, 55 categories | Large-scale 3D shape repository (Chang et al.) |
| ModelNet | 2015 | 12K CAD models, 40 categories | Princeton 3D object benchmark (Wu et al.) |
| ABC | 2019 | 1M+ CAD models | Mechanical parts with parametric annotations (Koch et al.) |
| Thingi10K | 2016 | 10K printable models | Web-derived 3D printing models with diverse topology (Zhou and Jacobson) |
| PartNet | 2019 | 27K objects, 573K parts | Part-level object annotations for structural decomposition (Mo et al.) |
| Text2Shape | 2018 | 75K text-shape pairs | Paired text and shape corpus for language-conditioned 3D generation (Chen et al.) |
| GSO (Google Scanned Objects) | 2022 | 1K+ scans | Household objects with PBR materials (Downs et al.) |
| ABO (Amazon Berkeley Objects) | 2022 | 8K+ models | Product catalog with multi-view images (Collins et al.) |
| CO3D | 2021 | 1.5M frames, 19K objects | Multi-view real-capture object dataset for category-level reconstruction (Reizenstein et al.) |
| Objaverse | 2023 | 800K+ objects | Internet-scale 3D asset collection (Deitke et al.) |
| Objaverse-XL | 2024 | 10.2M objects | Extended internet-scale collection (Deitke et al.) |
| Dataset | Year | Scale | Description |
|---|---|---|---|
| FAUST | 2014 | 300 scans, 10 subjects | Real body scans with ground-truth correspondence (Bogo et al.) |
| RenderPeople | 2018 | 4.5K+ subjects | Commercially scanned textured human meshes for character production (RenderPeople) |
| AMASS | 2019 | Large-scale motion capture | Unified motion capture archive (Mahmood et al.) |
| CAPE | 2020 | 4D clothing | Clothed body scans with pose variation (Ma et al.) |
| THuman2.0 | 2021 | 526 high-res scans | Detailed textured human models (Yu et al.) |
| HuMMan | 2022 | 1K subjects | Multi-modal human dataset (Cai et al.) |
| HumanML3D | 2022 | 14.6K text-motion sequences | Text-aligned motion corpus for controllable human generation and evaluation (Guo et al.) |
| Motion-X | 2024 | Large-scale motion | Expressive whole-body motion dataset (Lin et al.) |
| Dataset | Year | Scale | Description |
|---|---|---|---|
| ScanNet | 2017 | 1,513 indoor scans | RGB-D reconstructions with annotations (Dai et al.) |
| ScanNet++ | 2023 | 460 scenes | Laser+DSLR indoor scans with material annotations (Yeshwanth et al.) |
| Matterport3D | 2017 | 90 buildings | Large-scale indoor environments (Chang et al.) |
| HM3D | 2021 | 1K buildings | Habitat-scale indoor scans with navigation annotations (Ramakrishnan et al.) |
| Structured3D | 2020 | 3.5K houses | Synthetic indoor scenes with layout and topology labels (Zheng et al.) |
| Hypersim | 2021 | 461 scenes, 77K images | Photorealistic synthetic scenes with material labels (Roberts et al.) |
| 3D-FRONT | 2021 | 18K rooms | Professionally designed indoor layouts (Fu et al.) |
| ProcTHOR | 2022 | Procedural houses | Infinitely scalable simulated interiors (Deitke et al.) |
| Infinigen | 2023 | Procedural nature | Photorealistic procedural generation of natural worlds (Raistrick et al.) |
| Infinigen Indoors | 2024 | Procedural interiors | Indoor extension of Infinigen (Raistrick et al.) |
No open issues yet, or sync has not completed.