Research map / 3D vision / Gaussian splatting
iLRM: An Iterative Large 3D Reconstruction Model
Gyeongjin Kang, Seungtae Nam, Seungkwon Yang, Xiangyu Sun, Sameh Khamis, Abdelrahman Mohamed, Eunbyung Park
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 |
|---|---|---|---|
| replace per-layer cross-attention with a single first-layer cross-attention (remove iterative refinement) (ablation) | iLRM · RE10K (4, H, F) | PSNR: 28.58 -> 29.24 | Improved |
| add resolution decoupling of viewpoint tokens and image tokens (ablation) | iLRM · RE10K (4, H, F) | PSNR: 28.47 -> 29.24 | Improved |
| add token uplifting before cross-attention (ablation) | iLRM · RE10K (4, H, F) | PSNR: 28.90 -> 29.24 | Improved |
| increase update layers from 3 to 12 (ablation) | iLRM · RE10K (4, H, F) | PSNR: 28.04 -> 29.24 | Improved |
| use quarter mini-batch cross-attention instead of full cross-attention (ablation) | iLRM (8, H, F) training · RE10K | PSNR / training cost: 30.39 -> 30.08 PSNR; memory 62.5 -> 39.0 GB | Mixed |
| increase number of input views from 2 to 8 (half-resolution viewpoint tokens) | iLRM · RE10K (256x256) | PSNR: 28.65 -> 31.57 | Improved |
| increase views from 11 to 24 at 256x448 | DepthSplat · DL3DV 256x448 | PSNR: 24.28 -> 22.37 | Hurt |
| increase views from 11 to 24 at 256x448 | iLRM · DL3DV 256x448 | PSNR: 26.99 -> 27.38 | Improved |
| use iLRM vs Long-LRM with 16 views | DL3DV undistorted 540x960 · DL3DV undistorted 540x960, 16 views | LPIPS: 0.292 -> 0.295 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- DepthSplat — DepthSplat: Connecting Gaussian Splatting and Depth (CVPR 2025) · better on 4 of 4 shared results
- pixelSplat — pixelSplat: 3D Gaussian Splats from Image Pairs for Scalable Generalizable 3D Reconstruction (CVPR 2024) · better on 1 of 1 shared result
- MVSplat — MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images (ECCV 2024) · better on 1 of 1 shared result
- GS-LRM — GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting (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
- LRM: Large Reconstruction Model for Single Image to 3D (ICLR 2024 · similarity 0.90)
- LRM-Zero: Training Large Reconstruction Models with Synthesized Data (NeurIPS 2024 · similarity 0.89)
- DGS-LRM: Real-Time Deformable 3D Gaussian Reconstruction From Monocular Videos (NeurIPS 2025 · similarity 0.89)
- LIRM: Large Inverse Rendering Model for Progressive Reconstruction of Shape, Materials and View-dependent Radiance Fields (CVPR 2025 · similarity 0.88)
- 4D-LRM: Large Space-Time Reconstruction Model From and To Any View at Any Time (NeurIPS 2025 · similarity 0.88)
- VastGaussian: Vast 3D Gaussians for Large Scene Reconstruction (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.
