Research map / Image generation / Fast diffusion sampling
Generative Neural Fields by Mixtures of Neural Implicit Functions
Tackgeun You, Mijeong Kim, Jungtaek Kim, Bohyung Han
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 mNIF rather than Functa | CelebA-HQ 64² image reconstruction · CelebA-HQ 64² | PSNR: 26.6 -> 34.5 | Improved |
| use mNIF (L) rather than DPF | CelebA-HQ 64² image generation · CelebA-HQ 64² | FID: 13.2 -> 13.2 | No clear effect |
| use mNIF rather than GEM | ShapeNet 64³ voxel generation · ShapeNet 64³ | coverage: 0.409 -> 0.437 | Improved |
| use mNIF rather than GEM | ShapeNet 64³ voxel reconstruction · ShapeNet 64³ | MSE: 0.0153 -> 0.0161 | Hurt |
| use mNIF rather than Functa | SRN Cars scene reconstruction · SRN Cars | PSNR: 24.2 -> 25.9 | Improved |
| project layer-specific mixture coefficients from a latent vector instead of using layer-specific coefficients directly (ablation) | mNIF on CelebA-HQ 64² · CelebA-HQ 64² test | PSNR: 24.35 -> 25.09 | Improved |
| use layer-specific rather than shared coefficients across layers (ablation) | mNIF on CelebA-HQ 64² · CelebA-HQ 64² test | PSNR: 22.11 -> 24.35 | Improved |
| increase the number of mixture bases (ablation) | mNIF on CelebA-HQ 64² · CelebA-HQ 64² test | PSNR: 21.84 -> 25.09 -> 25.74 -> 25.85 | Improved |
| increase latent dimension H (ablation) | mNIF on CelebA-HQ 64² · CelebA-HQ 64² test | PSNR: 25.35 -> 27.92 -> 29.45 | Improved |
| include second-order gradients in meta-learning (ablation) | mNIF context adaptation on CelebA-HQ 64² · CelebA-HQ 64² | PSNR: 22.68 -> 26.64 | Improved |
| increase meta-learning inner steps from 3 to 5 (ablation) | mNIF context adaptation on CelebA-HQ 64² · CelebA-HQ 64² | PSNR: 26.64 -> 26.77 | Improved |
| extend context adaptation training from 800 to 1200 epochs (ablation) | mNIF (L) on CelebA-HQ 64² · CelebA-HQ 64² | FID: 14.30 -> 15.25 | Hurt |
Closest papers by meaning
- Neural Diffusion Models (ICML 2024 · similarity 0.86)
- Boosting Generative Image Modeling via Joint Image-Feature Synthesis (NeurIPS 2025 · similarity 0.86)
- DDMI: Domain-agnostic Latent Diffusion Models for Synthesizing High-Quality Implicit Neural Representations (ICLR 2024 · similarity 0.86)
- Analysis of Learning a Flow-based Generative Model from Limited Sample Complexity (ICLR 2024 · similarity 0.86)
- Image Neural Field Diffusion Models (CVPR 2024 · similarity 0.86)
- Diffusion Models for Multi-Task Generative Modeling (ICLR 2024 · similarity 0.86)
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.
