Research map / 3D vision / Face and shape
LSRM: High-Fidelity Object-Centric Reconstruction via Scaled Context Windows
Zhengqin Li, Cheng Zhang, Jakob Engel, Dong Zhao
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 |
|---|---|---|---|
| scale context window with sparse stage (S=96) vs dense model (ablation) | LSRM · GSO NVS (16 views) | PSNR: 28.45 -> 32.72 | Improved |
| increase S_img and S_vol from 64 to 96 with routing (ablation) | LSRM · GSO NVS (16 views) | PSNR: 31.93 -> 33.08 | Improved |
| add 3D-aware block routing at S=96 (ablation) | LSRM · GSO NVS (16 views) | PSNR: 32.72 -> 33.08 | Improved |
| add 3D-aware block routing at S=64 (ablation) | LSRM · GSO NVS (16 views) | LPIPS: 0.045 -> 0.040 | Improved |
| add 3D-aware routing at S=96 (ablation) | LSRM · GSO NVS (16 views) | LPIPS: 0.032 -> 0.028 | Improved |
| reduce S_vol 64 to 48 vs reduce S_img 64 to 48 (ablation) | LSRM · GSO NVS (16 views) | PSNR: 31.68 vs 31.82 | Improved |
| LSRM vs LIRM with 16 views | LSRM · GSO NVS (16 views) | PSNR: 30.65 -> 33.08 | Improved |
| fewer input views (8 instead of 16) | LSRM · GSO NVS (16 views) | PSNR: 33.08 -> 32.46 | Mixed |
| LSRM vs LIRM on StanfordORB | LSRM · StanfordORB | LPIPS: 0.025 -> 0.022 | Improved |
| LSRM vs NeuralPBIR on StanfordORB PSNR | LSRM · StanfordORB | PSNR-H: 26.01 vs 25.47 | Hurt |
| LSRM vs LIRM on StanfordORB geometry | LSRM · StanfordORB | Chamfer distance: 0.38 -> 0.29 | Improved |
| LSRM (16 views) vs 3DGS with 16 views | LSRM · GSO subset | PSNR: 25.10 -> 34.17 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GS-LRM — GS-LRM: Large Reconstruction Model for 3D Gaussian Splatting (ECCV 2024) · better on 2 of 2 shared results
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
- LIRM: Large Inverse Rendering Model for Progressive Reconstruction of Shape, Materials and View-dependent Radiance Fields (CVPR 2025 · similarity 0.89)
- Sparse3D: Distilling Multiview-Consistent Diffusion for Object Reconstruction from Sparse Views (AAAI 2024 · similarity 0.89)
- SPAR3D: Stable Point-Aware Reconstruction of 3D Objects from Single Images (CVPR 2025 · similarity 0.89)
- MegaSynth: Scaling Up 3D Scene Reconstruction with Synthesized Data (CVPR 2025 · similarity 0.89)
- Decompositional Neural Scene Reconstruction with Generative Diffusion Prior (CVPR 2025 · similarity 0.88)
- LRM-Zero: Training Large Reconstruction Models with Synthesized Data (NeurIPS 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.
