Research map / 3D vision / Gaussian splatting
F⁴Splat: Feed-Forward Predictive Densification for Feed-Forward 3D Gaussian Splatting
Injae Kim, Chaehyeon Kim, Minseong Bae, Minseok Joo, Hyunwoo Kim
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 |
|---|---|---|---|
| add learned densification score instead of random allocation (ablation) | F4Splat · RE10K (24 views, 20% Gaussian budget) | PSNR: 24.68 -> 25.47 | Improved |
| add learned densification score instead of Sobel frequency heuristic (ablation) | F4Splat · RE10K (24 views, 20% Gaussian budget) | PSNR: 25.36 -> 25.47 | Improved |
| add level-wise Gaussian supervision during training (ablation) | F4Splat · RE10K (24 views, 20% Gaussian budget) | PSNR: 24.25 -> 25.47 | Improved |
| add scene-scale regularization loss (ablation) | F4Splat · RE10K (24 views, 20% Gaussian budget) | PSNR: 4.82 -> 25.47 | Improved |
| use novel-view training with aligned target camera and scene-scale reg instead of AnySplat-like input-view training with frozen VGGT distillation (ablation) | F4Splat · VGGT-prior training comparison | PSNR: 20.59 -> 24.62 | Improved |
| same VGGT-prior strategy change on training time (ablation) | F4Splat · VGGT-prior training comparison | training time: 5H 35M -> 3H 03M | Improved |
| use F4Splat tau- (447K Gaussians) instead of AnySplat | F4Splat · RE10K (8 views) | PSNR: 24.07 -> 25.73 | Improved |
| use F4Splat tau+ with far fewer Gaussians instead of AnySplat | F4Splat · RE10K (24 views) | PSNR: 25.40 -> 26.01 | Improved |
| use F4Splat instead of NoPoSplat when trained on RE10K+DL3DV | F4Splat · DTU (2 views) | PSNR: 17.51 -> 19.76 | Improved |
| use F4Splat tau- instead of VicaSplat uncalibrated | F4Splat · ACID (2 views) | PSNR: 24.55 -> 26.26 | Improved |
| add adaptive allocation over uniform allocation (compute cost) | F4Splat · Computational overhead | inference time: 0.440 -> 0.488 | Hurt |
| use F4Splat pose estimation over VGGT | F4Splat · RE10K relative pose AUC@5 | AUC: 0.335 -> 0.541 | Improved |
Closest papers by meaning
- Reframing Gaussian Splatting Densification with Complexity-Density Consistency of Primitives (NeurIPS 2025 · similarity 0.93)
- CompactSplat: Spatially Adaptive Gaussian Distribution for Feedforward Scene Reconstruction (NeurIPS 2026 · similarity 0.92)
- Speedy-Splat: Fast 3D Gaussian Splatting with Sparse Pixels and Sparse Primitives (CVPR 2025 · similarity 0.92)
- Generative Sparse-View Gaussian Splatting (CVPR 2025 · similarity 0.91)
- Compressed 3D Gaussian Splatting for Accelerated Novel View Synthesis (CVPR 2024 · similarity 0.91)
- Swift4D: Adaptive divide-and-conquer Gaussian Splatting for compact and efficient reconstruction of dynamic scene (ICLR 2025 · similarity 0.91)
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.
