Research map / 3D vision / 3D generation
LN3Diff: Scalable Latent Neural Fields Diffusion for Speedy 3D Generation
Yushi Lan, Fangzhou Hong, Shuai Yang, Shangchen Zhou, Xuyi Meng, Bo Dai, Xingang Pan, Chen Change Loy
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 |
|---|---|---|---|
| replace 2D conv decoder with ViT block (ablation) | LN3Diff++ VAE · Objaverse subset 7K | PSNR@100K: 17.46 -> 18.92 | Improved |
| use DiT block with latent injected via adaLN instead of ViT block (ablation) | LN3Diff++ VAE decoder · Objaverse subset 7K | PSNR@100K: 18.92 -> 20.61 | Improved |
| add Plucker camera embedding to encoder input (ablation) | LN3Diff++ VAE · Objaverse subset 7K | PSNR@100K: 20.61 -> 21.29 | Improved |
| add cross-plane attention (ablation) | LN3Diff++ VAE decoder · Objaverse subset 7K | PSNR@100K: 21.29 -> 21.70 | Improved |
| add self-plane attention (ablation) | LN3Diff++ VAE decoder · Objaverse subset 7K | PSNR@100K: 21.70 -> 21.95 | Improved |
| add novel-view discriminator for monocular data (ablation) | LN3Diff++ VAE · FFHQ | novel-view plausibility (qualitative) | Improved |
| add DINO features to CLIP for image conditioning (ablation) | LN3Diff++ image-conditioned diffusion · Objaverse | faithfulness (qualitative) | Improved |
| diffuse in compact VAE latent space instead of latent-free neural fields | LN3Diff++ · Sampling speed, 1 V100 | seconds per instance: 5.7 vs 8.1 (SSDNeRF), 15.8 (RenderDiffusion), 18.7 (DiffRF) | Improved |
| train with only V=2 views instead of 50 views | LN3Diff++ vs SSDNeRF · ShapeNet Car COV | COV (%): 43.12 vs 67.82 (SSDNeRF V=50) | Hurt |
| fine-tune VAE with Flexicubes SDF for mesh extraction | LN3Diff++ · Objaverse image-to-3D | FID: 23.01 -> 36.28 | Hurt |
| reduce GAN weight of novel-view discriminator to limit background artifacts | LN3Diff++ on FFHQ · FFHQ | 3D plausibility | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- GaussianCube-V1.1 — GaussianCube: A Structured and Explicit Radiance Representation for 3D Generative Modeling (NeurIPS 2024) · better on 1 of 2 shared results
- DiffRF — DiffRF: Rendering-Guided 3D Radiance Field Diffusion (CVPR 2023) · better on 1 of 1 shared result
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
- WildFusion: Learning 3D-Aware Latent Diffusion Models in View Space (ICLR 2024 · similarity 0.90)
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 2025 · similarity 0.89)
- LaGeM: A Large Geometry Model for 3D Representation Learning and Diffusion (ICLR 2025 · similarity 0.88)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · similarity 0.88)
- Efficient-3Dim: Learning a Generalizable Single-image Novel-view Synthesizer in One Day (ICLR 2024 · similarity 0.88)
- Denoising Diffusion via Image-Based Rendering (ICLR 2024 · similarity 0.87)
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.
