Research map / 3D vision / Face and shape
3D shape generative modeling
Generates 3D shapes with deep generative models, latent grids, primitives and text conditioning.
74 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Face and shape, keyword cluster “3d shape · shapes · object”
State History Tried Baselines Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
17 papers in 2023–24 and 25 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 4.7% to 6.8%.
Open questions the data points to
- Papers report Liu et al. on Text2Shape chairs (CLIP-S) anywhere from 29.94 to 38.88: comparisons against it depend on whose number you take.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling (NeurIPS 2016 · 1,544 citations)
- 2017 · Synthesizing 3D Shapes via Modeling Multi-View Depth Maps and Silhouettes With Deep Generative Networks (CVPR 2017 · 202 citations)
- 2018 · A Papier-Mâché Approach to Learning 3D Surface Generation (CVPR 2018 · 953 citations)
- 2019 · Learning Implicit Fields for Generative Shape Modeling (CVPR 2019 · 1,503 citations)
- 2020 · BSP-Net: Generating Compact Meshes via Binary Space Partitioning (CVPR 2020 · 290 citations)
- 2021 · DECOR-GAN: 3D Shape Detailization by Conditional Refinement (CVPR 2021 · 48 citations)
- 2022 · Efficient Geometry-Aware 3D Generative Adversarial Networks (CVPR 2022 · 1,071 citations)
- 2023 · SDFusion: Multimodal 3D Shape Completion, Reconstruction, and Generation (CVPR 2023 · 201 citations)
- 2024 · Mosaic-SDF for 3D Generative Models (CVPR 2024 · 19 citations)
- 2025 · Dora: Sampling and Benchmarking for 3D Shape Variational Auto-Encoders (CVPR 2025 · 20 citations)
- 2026 · Native and Compact Structured Latents for 3D Generation (CVPR 2026 · 4 citations)
Earliest papers here
- Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling (NeurIPS 2016 · 1,544 citations)
- Synthesizing 3D Shapes via Modeling Multi-View Depth Maps and Silhouettes With Deep Generative Networks (CVPR 2017 · 202 citations)
- SurfNet: Generating 3D Shape Surfaces Using Deep Residual Networks (CVPR 2017 · 176 citations)
What has been tried
56 papers here report 566 measured changes (a component added, removed or swapped, compared with a baseline): helped 72%, no clear effect 3%, hurt 21%, mixed 4%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add sharp edge sampling and dual cross-attention | Dora-bench L4 · F-score (0.005) × 100 | 68.669 -> 87.473 | Helped | Dora: Sampling and Benchmarking for 3D Shape Variational Auto-Encoders (CVPR 2025 · 20 citations) |
| remove latent alignment | AmateurSketch-3D · CD | 0.109 -> 0.171 | Hurt | Doodle Your 3D: From Abstract Freehand Sketches to Precise 3D Shapes (CVPR 2024 · 16 citations) |
| Use SDXL-inpainting rather than Instant3D as the fine-tuning backbone | Multiview text-to-image inpainting benchmark · ClipG | 42.50 -> 43.48 | Helped | Instant3dit: Multiview Inpainting for Fast Editing of 3D Objects (CVPR 2025 · 14 citations) |
| Add 3D merging to the multi-view editing pipeline | Edited-to-input shape consistency · Chamfer Distance | 2.95 -> 2.28 | Helped | PrEditor3D: Fast and Precise 3D Shape Editing (CVPR 2025 · 9 citations) |
| Predict edit direction and magnitude separately instead of a coupled edit vector | ShapeWalk chains |P| ∈ {10, 15, 20} · FCD×1e4 | 2.990 -> 2.670 | Helped | ShapeWalk: Compositional Shape Editing Through Language-Guided Chains (CVPR 2024 · 6 citations) |
| Add the geometric field loss | Emoji, Icon, and Clipart · MSE | 0.0013 -> 0.0008; 0.0018 -> 0.0011; 0.0043 -> 0.0013 | Helped | Towards High-fidelity Artistic Image Vectorization via Texture-Encapsulated Shape Parameterization (CVPR 2024 · 6 citations) |
| add triplane-aware convolution to the denoising network | 10 testing examples (Avg.) · G-Qual. | 0.347 -> 0.156 | Helped | Sin3DM: Learning a Diffusion Model from a Single 3D Textured Shape (ICLR 2024 · 6 citations) |
| add the hyperbolic text-image encoder (HTIE) | Text2Shape mini-set · IoU | 12.74 -> 14.26 | Helped | HyperSDFusion: Bridging Hierarchical Structures in Language and Geometry for Enhanced 3D Text2Shape Generation (CVPR 2024 · 5 citations) |
| Use shape-conditioning instead of no shape-conditioning | HumanML3D reconstruction · Bone Length Diff. (mm) | 99.18 -> 66.41 | Helped | Shape My Moves: Text-Driven Shape-Aware Synthesis of Human Motions (CVPR 2025 · 4 citations) |
| use neural deformation instead of topology fixing and Tutte embedding | VDM reconstruction · CLIPImg | 0.8401 -> 0.8521 | Helped | GenVDM: Generating Vector Displacement Maps From a Single Image (CVPR 2025 · 3 citations) |
| Increase shape guidance strength λ | Shape guidance examples · Shape resemblance | Helped | ShapeWords: Guiding Text-to-Image Synthesis with 3D Shape-Aware Prompts (CVPR 2025 · 2 citations) | |
| replace CIPS-UV with CIPS-2D texture mapping | CompCars · FID | 41.79 -> 148.09 | Hurt | TUVF: Learning Generalizable Texture UV Radiance Fields (ICLR 2024 · 2 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| Liu et al. | Text2Shape chairs | CLIP-S | 29.94 – 38.88 | 2 papers, e.g. CVPR 2023 and NeurIPS 2022 |
What reviewers will expect
Datasets and baselines named in the 17 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Objaverse · 7 of 17 papers
- ShapeNet · 4 of 17 papers
- GSO · 2 of 17 papers
- ABO · 2 of 17 papers
Baselines
- 3DShape2VecSet · 2 of 17 papers
- Wonder3D · 2 of 17 papers
- Vox-E · 2 of 17 papers
- MVEdit · 2 of 17 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 42 papers; from OpenAlex and the papers' affiliations).
- Adobe · 3 papers
- University of Illinois, Urbana Champaign · 2 papers
- Technical University of Munich · 2 papers
- Meta · 2 papers
- KAUST · 2 papers
- University of Texas at Austin · 2 papers
- Snap Inc · 2 papers
- The Chinese University of Hong Kong · 2 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Primitive-based shape abstraction (99 papers)
- Single-view 3D reconstruction (96 papers)
- Articulated and non-rigid shape learning (75 papers)
- Optimization-based shape matching (59 papers)
Elsewhere in the topic
- Neural implicit surface reconstruction (131 papers)
- Gaussian and implicit head avatars (115 papers)
- Textured mesh generation (111 papers)
- Face reflectance and relightable rendering (85 papers)
- 3D morphable face model regression (80 papers)
- Full-body avatar reconstruction (78 papers)
- Learned mesh and geometric representations (73 papers)
- Clothed human modeling (60 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).