Research map / 3D vision / 3D generation
StructLDM: Structured Latent Diffusion for 3D Human Generation
Tao Hu, Fangzhou Hong, Ziwei Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use structure-aligned normalization in latent diffusion instead of none (ablation) | StructLDM diffusion · UBCFashion | FID@4K: 20.13 -> 19.58 | Improved |
| use standard normalization in diffusion (ablation) | StructLDM diffusion · UBCFashion | FID@4K: 20.13 -> 24.86 | Hurt |
| use part-aware normalization in diffusion (ablation) | StructLDM diffusion · UBCFashion | FID@4K: 20.13 -> 25.02 | Hurt |
| use structured 2D latent instead of global 1D latent (ablation) | auto-decoder · UBCFashion reconstruction | LPIPS | Improved |
| increase 2D latent resolution or channels (ablation) | auto-decoder · UBCFashion reconstruction | LPIPS | Improved |
| increase 2D latent from 64x64x32 to 128x128x16 (ablation) | auto-decoder · UBCFashion reconstruction | LPIPS | No clear effect |
| add adversarial training to auto-decoder (ablation) | auto-decoder · UBCFashion qualitative | reconstruction quality | Improved |
| add latent noising and denoising (Diff-Render) to compositional blending (ablation) | StructLDM · compositional generation | color blending quality | Improved |
| add mapping network to smooth latents on single-view DeepFashion (ablation) | auto-decoder · DeepFashion | generation quality vs reconstruction fidelity | Mixed |
| StructLDM versus EVA3D on THUman2.0 | StructLDM · THUman2.0 | FID@50K: 124.54 -> 25.22 | Improved |
| StructLDM on single-view DeepFashion versus AG3D | StructLDM · DeepFashion | FID: 10.93 -> 20.82 | Hurt |
Later papers that compared with this method
1 later accepted paper reports StructLDM as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Disentangled Clothed Avatar Generation with Layered Representation (ICCV 2025 · better on 1 of 1 shared result)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- EVA3D — EVA3D: Compositional 3D Human Generation from 2D Image Collections (ICLR 2023) · better on 2 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- HyperHuman: Hyper-Realistic Human Generation with Latent Structural Diffusion (ICLR 2024 · similarity 0.91)
- HumanNorm: Learning Normal Diffusion Model for High-quality and Realistic 3D Human Generation (CVPR 2024 · similarity 0.90)
- SyncHuman: Synchronizing 2D and 3D Generative Models for Single-view Human Reconstruction (NeurIPS 2025 · similarity 0.89)
- Joint2Human: High-Quality 3D Human Generation via Compact Spherical Embedding of 3D Joints (CVPR 2024 · similarity 0.89)
- Human-3Diffusion: Realistic Avatar Creation via Explicit 3D Consistent Diffusion Models (NeurIPS 2024 · similarity 0.89)
- SiTH: Single-view Textured Human Reconstruction with Image-Conditioned Diffusion (CVPR 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
