Research map / Point clouds / Shape and surface completion
Learning Gradient Fields for Shape Generation
Ruojin Cai, Guandao Yang, Hadar Averbuch-Elor, Zekun Hao, Serge Belongie, Noah Snavely, Bharath Hariharan
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 |
|---|---|---|---|
| train with multiple noise levels instead of a single level 0.1 (ablation) | gradient field model · Airplane auto-encoding | CD (x1e4): 2.545 -> 0.993 | Improved |
| train with multiple noise levels instead of a single level 0.05 (ablation) | gradient field model · Airplane auto-encoding | CD (x1e4): 1.573 -> 0.993 | Improved |
| train with multiple noise levels instead of a single level 0.01 (ablation) | gradient field model · Airplane auto-encoding | CD (x1e4): 1009.357 -> 0.993 | Improved |
| initialize prior with Gaussian or fixed point instead of uniform (ablation) | gradient field model · Airplane auto-encoding | CD (x1e4): 0.993 -> 0.996 | No clear effect |
| add shortcuts, CBN latent input and decoder concat on top of (a) (ablation) | architecture · Airplane auto-encoding | CD (x1e4): 1.234 -> 0.987 | Improved |
| add shortcuts to CBN variant (ab vs a) (ablation) | architecture · Airplane auto-encoding | CD (x1e4): 1.234 -> 0.992 | Improved |
| use latent GAN instead of VAE for shape distribution (ablation) | gradient field model · Airplane generation | MMD-CD (x1e3): 1.909 -> 1.332 | Improved |
| use latent GAN instead of VAE for shape distribution (ablation) | gradient field model · Chair generation | MMD-CD (x1e3): 18.032 -> 14.818 | Improved |
| reduce input points M from 2048 to 128 (ablation) | gradient field model · Airplane sparse auto-encoding (N=10K) | CD (x1e4): 0.993 -> 1.688 | Hurt |
| reduce number of available points N from 10K to 1K | gradient field model · Airplane sparse auto-encoding (M=512) | CD (x1e4): 0.999 -> 1.021 | No clear effect |
| use gradient field model instead of PointFlow | gradient field model · Chair generation | COV-CD (%): 40.94 -> 46.37 | Improved |
| use gradient field model instead of PointFlow | gradient field model · Airplane generation | 1-NNA-CD (%): 83.21 -> 85.06 | Hurt |
Closest papers by meaning
- Diffusion Probabilistic Models for 3D Point Cloud Generation (CVPR 2021 · similarity 0.90)
- PointNSP: Autoregressive 3D Point Cloud Generation with Next-Scale Level-of-Detail Prediction (CVPR 2026 · similarity 0.90)
- Learning Progressive Point Embeddings for 3D Point Cloud Generation (CVPR 2021 · similarity 0.90)
- Autoregressive 3D Shape Generation via Canonical Mapping (ECCV 2022 · similarity 0.90)
- PointGMM: A Neural GMM Network for Point Clouds (CVPR 2020 · similarity 0.89)
- Point Cloud Pre-training with Diffusion Models (CVPR 2024 · similarity 0.88)
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.
