Research map / 3D vision / Face and shape
DNF: Unconditional 4D Generation with Dictionary-based Neural Fields
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes DNF, a dictionary-based neural field representation for 4D deforming shapes (SVD-derived shared dictionary plus per-shape latents and coefficients), and trains a transformer diffusion model on it for unconditional 4D generation.
Problem
Unconditional generative modeling of 4D deforming shapes, where existing neural field approaches either lose shape detail when using a single global latent or lack shared structure when fitted per shape.
Approach
Decompose pre-trained shape and motion MLPs with SVD into a shared dictionary of singular vectors, fine-tune only per-shape singular-value coefficients, compress and extend the dictionary with low-rank orthogonal residuals, and then run a transformer diffusion model over the coefficients and latent codes.
Key results
- Experiments on DeformingThings4D demonstrate the effectiveness of the approach, achieving state-of-the-art unconditional generation of deforming objects (per the paper's claims).
Limitations
- Evaluation text provided is truncated; quantitative results not visible.
Experimental setup
Datasets: DeformingThings4D
Benchmarks: DeformingThings4D
Baselines: HyperDiffusion Motion2VecSets
Builds on: Neural Parametric Models HyperDiffusion Motion2VecSets DeepSDF
Closest papers by meaning
- Sculpt4D: Generating 4D Shapes via Sparse-Attention Diffusion Transformers (CVPR 2026 · similarity 0.88)
- Mesh4D: 4D Mesh Reconstruction and Tracking from Monocular Video (CVPR 2026 · similarity 0.87)
- Motion2VecSets: 4D Latent Vector Set Diffusion for Non-rigid Shape Reconstruction and Tracking (CVPR 2024 · similarity 0.86)
- A Unified Approach for Text- and Image-guided 4D Scene Generation (CVPR 2024 · similarity 0.86)
- Category-Agnostic Neural Object Rigging (CVPR 2025 · similarity 0.85)
- Disco4D: Disentangled 4D Human Generation and Animation from a Single Image (CVPR 2025 · similarity 0.85)
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.