Research map / 3D vision / Gaussian splatting
iSplat: Iterative Learning for Fine-Grained Gaussian Splatting
Haifeng Wu, Wei Long, Shuhang Gu, Lixin Duan, Wen Li
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 iterative GRU-based learning (3 passes) over single pass (ablation) | iSplat · RealEstate10K | PSNR: 26.95 -> 27.67 | Improved |
| add uncertainty-driven depth refinement instead of static range (ablation) | iSplat · RealEstate10K | PSNR: 27.10 -> 27.67 | Improved |
| add region-aware feature enhancement (ablation) | iSplat · RealEstate10K | PSNR: 27.42 -> 27.67 | Improved |
| increase iterations from 1 to 3 (train and test) (ablation) | iSplat · RealEstate10K | PSNR: 26.95 -> 27.60 -> 27.67 | Improved |
| increase iterations from 3 to 4 (ablation) | iSplat · RealEstate10K | PSNR: 27.67 -> 27.65 | No clear effect |
| mismatch training and test iteration counts (ablation) | iSplat · RealEstate10K | PSNR: 27.67 (3->3) vs 27.58 (3->1) | Hurt |
| use iSplat (42.6M) instead of larger DepthSplat-L (354M) | feed-forward 3DGS · RealEstate10K | PSNR: 27.47 -> 27.67 | Improved |
| use iSplat instead of DepthSplat for zero-shot transfer | feed-forward 3DGS · Re10k to DTU | PSNR: 15.38 -> 18.26 | Improved |
| use iSplat instead of DepthSplat for zero-shot transfer | feed-forward 3DGS · Re10k to ACID | PSNR: 28.37 -> 28.65 | Improved |
| add 3 iterative passes versus DepthSplat in inference cost | iSplat · RealEstate10K | inference time (s): 0.098 -> 0.124 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- DepthSplat — DepthSplat: Connecting Gaussian Splatting and Depth (CVPR 2025) · better on 3 of 3 shared results
- MonoSplat — MonoSplat: Generalizable 3D Gaussian Splatting from Monocular Depth Foundation Models (CVPR 2025) · better on 3 of 3 shared results
- MVSplat — MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images (ECCV 2024) · 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
- DepthSplat: Connecting Gaussian Splatting and Depth (CVPR 2025 · similarity 0.92)
- HiSplat: Hierarchical 3D Gaussian Splatting for Generalizable Sparse-View Reconstruction (ICLR 2025 · similarity 0.92)
- TranSplat: Generalizable 3D Gaussian Splatting from Sparse Multi-View Images with Transformers (AAAI 2025 · similarity 0.91)
- pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction (CVPR 2024 · similarity 0.91)
- GaussianSpa: An "Optimizing-Sparsifying" Simplification Framework for Compact and High-Quality 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- Learning Efficient Fuse-and-Refine for Feed-Forward 3D Gaussian Splatting (NeurIPS 2025 · similarity 0.90)
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.
